{"meta":{"title":"Frontier Model Bench","version":"0.3.0-derived","dataVersion":"canonical-catalog-v1","asOf":"2026-08-28","status":"curated","lastUpdated":"2026-08-28T00:00:00Z","owner":"Cunxin Huang","note":"Canonical catalog + long-form observations；公开披露层默认作为 coverage map 展示，catalog-only 行无分数且不参与成绩计算。","updateCadence":"每周复核，重大模型发布时加急","defaultView":"atlas","defaultPreset":"public-coverage","defaultBenchmarkIds":["gpqa-diamond","aime-2025","aime-2026","mmlu-pro","hle","swebench-verified","swebench-pro","livecodebench","deepswe","terminal-bench","mmmu","mmmu-pro"],"generatedFrom":["data/catalog/models.json","data/catalog/benchmarks.json","data/catalog/benchmark_profiles.json","data/catalog/model_profiles.json","data/catalog/sources.json","data/catalog/harnesses.json","data/catalog/presets.json","data/observations/results.jsonl","data/models.json (legacy fallback)","data/derived/public.json (reported/unverified display layer, optional)"]},"benchmarks":[{"id":"gpqa-diamond","canonicalId":"gpqa-diamond","short":"GPQA","name":"GPQA Diamond","family":"reasoning","category":"reasoning","familyLabel":"推理","evaluationMode":"direct","comparisonKey":"gpqa-diamond:gpqa-diamond@v1:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"Diamond","versionId":"gpqa-diamond@v1","description":"研究生级别的生物、物理与化学问答；题目由领域专家编写。","owner":"Stanford / Scale","source":"https://arxiv.org/abs/2311.12022","sourceLabel":"GPQA paper","sourceIds":["gpqa-paper"],"lastVerified":"2026-08-28","versions":[{"id":"gpqa-diamond@v1","label":"Diamond","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":2,"profile":{"task":{"summary":"研究生级生物、物理、化学多项选择题，要求在领域知识不足以直接检索的条件下完成推理。","input":"题干与选项（通常四选一）","output":"选项答案"},"dataset":{"size":"Diamond 198 题；GPQA 全集 448 题","splits":["Diamond"],"availability":"公开数据集；答案键用于离线判分"},"protocol":{"mode":"model-only / direct","grader":"答案 exact-match accuracy","tools":"通常 closed-book、无浏览；若启用检索需单独标注","sampling":"temperature、shot 和答案解析方式需披露"},"comparison":{"recommended":"同一 Diamond 版本、prompt、shot 和解码设置下比较 STEM 推理","avoid":"不要把 GPQA 全集、Diamond、带搜索和带 agent 的结果混成一列"},"source_locator":"GPQA paper §3 与数据集说明；Diamond 子集定义"}},{"id":"aime-2025","canonicalId":"aime-2025","short":"AIME","name":"AIME 2025","family":"reasoning","category":"math","familyLabel":"数学","evaluationMode":"direct","comparisonKey":"aime-2025:aime-2025@v1:pass@1","metric":"pass@1","metricLabel":"Pass@1","scale":100,"unit":"%","direction":"higher","version":"2025","versionId":"aime-2025@v1","description":"数学竞赛题；采样次数、工具和答案抽取方式必须明确。","owner":"AIME","source":"https://maa.org/maa-invitational-competitions/","sourceLabel":"MAA · American Invitational Mathematics Examination","sourceIds":["src-maa-aime","openai-gpt5"],"lastVerified":"2026-08-28","versions":[{"id":"aime-2025@v1","label":"2025","status":"active"}],"metrics":[{"id":"pass@1","label":"Pass@1","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":10,"profile":{"task":{"summary":"AIME 竞赛数学整数答案题，考察多步推导、计算和最终答案提取。","input":"竞赛题文本","output":"0–999 整数答案"},"dataset":{"size":"2025 年 AIME I/II 题目窗口（每场 15 题；总量以发布快照为准）","splits":["AIME I 2025","AIME II 2025"],"availability":"题目公开；答案与评测脚本按来源快照记录"},"protocol":{"mode":"model-only / direct","grader":"整数 exact match","tools":"是否允许 Python/计算器必须注明","sampling":"pass@1、majority vote、best-of-n 不可互换"},"comparison":{"recommended":"固定年份、题目窗口、采样预算和答案抽取规则","avoid":"不要用 best-of-n 或泄露题解的成绩冒充单次 pass@1"},"source_locator":"MAA Invitational Competitions official AIME source + OpenAI GPT-5 evaluation table；保留 AIME I/II、30题聚合、工具、n 与题解公开截止时间"}},{"id":"aime-2026","canonicalId":"aime-2026","short":"AIME 2026","name":"AIME 2026","family":"reasoning","category":"math","familyLabel":"数学","evaluationMode":"direct","comparisonKey":"aime-2026:aime-2026@v1:pass@1","metric":"pass@1","metricLabel":"Pass@1","scale":100,"unit":"%","direction":"higher","version":"2026","versionId":"aime-2026@v1","description":"AIME 2026 题集；题目窗口、采样次数、工具和答案抽取方式必须明确。","owner":"MathArena / AIME","source":"https://maa.org/maa-invitational-competitions/","sourceLabel":"MAA · American Invitational Mathematics Examination","sourceIds":["src-maa-aime","hf-aime-2026"],"lastVerified":"2026-08-28","versions":[{"id":"aime-2026@v1","label":"2026","status":"active"}],"metrics":[{"id":"pass@1","label":"Pass@1","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":9,"profile":{"task":{"summary":"最新 AIME 数学整数答案题，用于观察前沿模型在新题窗口上的竞赛数学能力。","input":"AIME 2026 题目文本","output":"0–999 整数答案"},"dataset":{"size":"AIME I/II 各 15 题；combined 口径为 30 题","splits":["AIME I 2026","AIME II 2026","combined 30"],"availability":"MAA competition source + MathArena/HF 公开榜单快照"},"protocol":{"mode":"model-only / direct","grader":"整数 exact match","tools":"是否使用代码执行、搜索或外部证明需记录","sampling":"记录 temperature、n、答案归一化和截止日期"},"comparison":{"recommended":"把同一题集快照与相同 inference budget 的模型并列","avoid":"不要把 2025 与 2026 或公开题解之后的回测直接平均"},"source_locator":"MAA Invitational Competitions official AIME source + MathArena AIME 2026 leaderboard API；HMMT 是不同竞赛，不作为 AIME 来源"}},{"id":"frontiermath","canonicalId":"frontiermath","short":"FrontierMath","name":"FrontierMath T1–T3","family":"reasoning","category":"math","familyLabel":"数学","evaluationMode":"direct","comparisonKey":"frontiermath:frontiermath@v2-t1-t3:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"v2 · Tiers 1–3 · 295 problems","versionId":"frontiermath@v2-t1-t3","description":"FrontierMath v2 的 Tiers 1–3（295 题）；v2 修正了大量题目，必须与 v1 分开并保留 Python/tool/token 协议。","owner":"Epoch AI","source":"https://epoch.ai/benchmarks/frontiermath","sourceLabel":"FrontierMath benchmark","sourceIds":["src-frontiermath","src-frontiermath-v2"],"lastVerified":"2026-08-28","versions":[{"id":"frontiermath@t1-t3","label":"T1–T3 pre-v2","status":"legacy"},{"id":"frontiermath@v2-t1-t3","label":"v2 · Tiers 1–3 · 295 problems","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"面向前沿研究数学的开放式/短答案问题，按难度层级测试模型能否提出并验证非模板化解法。","input":"研究级数学问题与必要定义","output":"可判定的数学答案或证明"},"dataset":{"size":"FrontierMath v2 共 338 题：Tiers 1–3 为 295 题，Tier 4 为 43 题；v2 修正了 42% 题目","splits":["v2 Tier 1","v2 Tier 2","v2 Tier 3"],"availability":"部分题目/答案受限；v1/v2 结果必须分开"},"protocol":{"mode":"model-only 或带工具的 system，须分开","grader":"官方 answer() grader / 专家规则，grader version 随 release 固定","tools":"Python tool、CAS、检索和 proof checker 需逐项记录","sampling":"v2 官方 protocol 最高 1M tokens；pass@1、多次采样与 partial credit 需区分"},"comparison":{"recommended":"同一 tier、题目窗口、工具许可和 judge 版本内比较","avoid":"不要将 T1–T3 与更难 tier 或不同保密窗口合并为一个总分"},"source_locator":"Epoch AI FrontierMath Tier 4 v2 official page（2026-06-12 release）；v2 task counts、42% corrections、Python tool、1M-token limit 与 answer() grader"}},{"id":"hmmt","canonicalId":"hmmt","short":"HMMT","name":"HMMT","family":"reasoning","category":"math","familyLabel":"数学","evaluationMode":"direct","comparisonKey":"hmmt:hmmt@2026:pass@1","metric":"pass@1","metricLabel":"Pass@1","scale":100,"unit":"%","direction":"higher","version":"February 2026","versionId":"hmmt@2026","description":"数学竞赛题；应区分年份、赛季和 avg@k 设置。","owner":"HMMT","source":"https://www.hmmt.org/","sourceLabel":"HMMT benchmark","sourceIds":["src-hmmt","hf-hmmt-2026"],"lastVerified":"2026-08-28","versions":[{"id":"hmmt@2025","label":"2025","status":"legacy"},{"id":"hmmt@2026","label":"February 2026","status":"active"}],"metrics":[{"id":"pass@1","label":"Pass@1","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"HMMT 竞赛数学题，覆盖代数、几何、组合和数论，测试精确解题与答案提取。","input":"HMMT 赛季题目","output":"短答案/整数或选择项"},"dataset":{"size":"按比赛月份、年份和科目分组；题量随赛季发布","splits":["February","November","subject/round slices"],"availability":"题目与公开评测结果按赛季保存"},"protocol":{"mode":"model-only / direct","grader":"答案 exact match 或赛季官方规则","tools":"是否允许代码、查表和多轮自检需披露","sampling":"pass@1、avg@k、majority vote 分开记录"},"comparison":{"recommended":"同一年份/赛季、科目、shot 和采样预算内比较","avoid":"不要跨赛季直接平均，也不要把 avg@k 当单次 pass@1"},"source_locator":"HMMT 官方竞赛页面与 MathArena/HF 赛季数据卡"}},{"id":"mmlu-pro","canonicalId":"mmlu-pro","short":"MMLU-Pro","name":"MMLU-Pro","family":"knowledge","category":"knowledge","familyLabel":"知识","evaluationMode":"direct","comparisonKey":"mmlu-pro:mmlu-pro@v1:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"v1","versionId":"mmlu-pro@v1","description":"更高难度、减少提示泄漏的多学科知识评测。","owner":"TIGER-AI-Lab","source":"https://github.com/TIGER-AI-Lab/MMLU-Pro","sourceLabel":"MMLU-Pro repository","sourceIds":["mmlu-pro"],"lastVerified":"2026-08-28","versions":[{"id":"mmlu-pro@v1","label":"v1","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":11,"profile":{"task":{"summary":"减少提示泄漏、扩大选项空间的多学科知识与推理选择题。","input":"题干、学科标签与十个候选项","output":"选项答案"},"dataset":{"size":"约 12K 题、14 个学科（以官方 release 为准）","splits":["test","subject slices"],"availability":"公开数据与评测代码"},"protocol":{"mode":"model-only / direct","grader":"选项 exact-match accuracy；总体通常按题目或学科聚合","tools":"默认无外部工具；工具增强结果另列","sampling":"shot、选项顺序、prompt 模板和答案解析需固定"},"comparison":{"recommended":"同一 release、同一 subject aggregation 与 prompt 下比较","avoid":"不要把 MMLU 原版、MMLU-Pro 和不同学科加权方式混合"},"source_locator":"TIGER-AI-Lab MMLU-Pro repository 与数据卡"}},{"id":"hle","canonicalId":"hle","short":"HLE","name":"Humanity's Last Exam","family":"knowledge","category":"knowledge","familyLabel":"知识","evaluationMode":"direct","comparisonKey":"hle:hle@v1:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"full","versionId":"hle@v1","description":"专家级跨学科问题；必须区分 text、vision、tools 和 judge 设置。","owner":"Scale AI","source":"https://labs.scale.com/leaderboard/humanitys_last_exam","sourceLabel":"Scale Labs HLE leaderboard","sourceIds":["hle"],"lastVerified":"2026-08-28","versions":[{"id":"hle@v1","label":"full","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":3,"profile":{"task":{"summary":"跨学科专家级闭卷问题，包含文本与多模态题，专门避免已饱和的简单知识问答。","input":"文本或图像问题（closed-ended）","output":"短答案/选项答案及可判定解释"},"dataset":{"size":"finalized public release：2,500 题、100+ 学科；另有 HLE-preview、text-only 与 private held-out scope","splits":["finalized public 2025-04-03","preview","text-only","vision","private held-out","subject slices"],"availability":"公开题集与 official leaderboard；private held-out 用于抗过拟合"},"protocol":{"mode":"model-only 或 tools-enabled system，必须标注","grader":"accuracy + CI 与 RMS calibration error；judge model/prompt/version 需保留","tools":"无工具、搜索和工具辅助是不同 track","sampling":"temperature、reasoning budget、图像输入、答案格式和 public/private scope 需记录"},"comparison":{"recommended":"同一 text/vision/tools track、prompt、judge 和题集版本内比较","avoid":"不要把工具辅助分数当作裸模型知识分，也不要忽略题目难度/subject mix"},"source_locator":"Scale Labs Humanity's Last Exam official leaderboard；finalized 2025-04-03 / preview / text-only / private holdout 分轨"}},{"id":"simpleqa","canonicalId":"simpleqa","short":"SimpleQA","name":"SimpleQA","family":"knowledge","category":"reliability","familyLabel":"事实性","evaluationMode":"direct","comparisonKey":"simpleqa:simpleqa@v1:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"v1","versionId":"simpleqa@v1","description":"短事实问答；建议同时保存正确率、拒答率和校准信息。","owner":"OpenAI","source":"https://github.com/openai/simple-evals","sourceLabel":"SimpleQA benchmark","sourceIds":["src-simpleqa"],"lastVerified":"2026-08-28","versions":[{"id":"simpleqa@v1","label":"v1","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"短事实问答，要求模型给出简洁、可核验且不过度扩展的事实。","input":"单句事实问题","output":"短文本答案或明确拒答"},"dataset":{"size":"4,326 个 fact-seeking questions","splits":["all","topic slices","SimpleQA-no-browse protocol"],"availability":"OpenAI simple-evals 公开"},"protocol":{"mode":"model-only / direct","grader":"reference match 与 grader 判断 correct/incorrect/ambiguous","tools":"no-browse 是标准可比设置；浏览版需另列","sampling":"同时记录 accuracy、拒答率、校准和 grader 模型"},"comparison":{"recommended":"固定 no-browse、知识截止日期、prompt 和 grader 版本","avoid":"不要只看 accuracy 而忽略 abstention/calibration，也不要混用 browsing 结果"},"source_locator":"OpenAI SimpleQA paper 与 simple-evals repository"}},{"id":"swebench-verified","canonicalId":"swebench-verified","short":"SWE-bench","name":"SWE-bench Verified","family":"coding","category":"coding-agent","familyLabel":"代码 Agent","evaluationMode":"system","comparisonKey":"swebench-verified:swebench-verified@verified:resolved","metric":"resolved","metricLabel":"Resolved","scale":100,"unit":"%","direction":"higher","version":"Verified","versionId":"swebench-verified@verified","description":"真实 GitHub issue 的端到端修复率；模型、提示词、工具和 harness 共同决定结果。","owner":"SWE-bench","source":"https://www.swebench.com/","sourceLabel":"SWE-bench official leaderboard","sourceIds":["swebench"],"lastVerified":"2026-08-28","versions":[{"id":"swebench-verified@verified","label":"Verified","status":"active"}],"metrics":[{"id":"resolved","label":"Resolved","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":5,"profile":{"task":{"summary":"从真实 GitHub issue、仓库上下文到补丁和测试通过的端到端软件工程任务。","input":"issue 描述、指定 commit 的代码仓库","output":"可应用的代码 patch"},"dataset":{"size":"500 个 human-filtered verified instances","splits":["Verified","repository/project slices"],"availability":"任务与评测 harness 公开"},"protocol":{"mode":"system / agent harness","grader":"应用 patch 后运行测试，报告 Resolved %","tools":"official full-system leaderboard 与 bash-only mini-SWE-agent 是不同 track；终端、编辑、网络和 checked 状态必须注明","sampling":"rollout 数、patch selection、timeout、mini-SWE-agent 1.x/2.x 与 scaffold version 影响结果"},"comparison":{"recommended":"优先比较相同 harness（如 mini-SWE-agent）、任务版本和预算","avoid":"不要把 agent/system 分数标成裸模型；注意公开 issue/patch 训练污染风险"},"source_locator":"swebench.com/verified.html + official leaderboard；保留 track、mini-SWE-agent release、scaffold、checked、cost 与 trajectory"}},{"id":"swebench-pro","canonicalId":"swebench-pro","short":"SWE-Pro","name":"SWE-bench Pro","family":"coding","category":"coding-agent","familyLabel":"代码 Agent","evaluationMode":"system","comparisonKey":"swebench-pro:swebench-pro@v1:resolved","metric":"resolved","metricLabel":"Resolved","scale":100,"unit":"%","direction":"higher","version":"v1","versionId":"swebench-pro@v1","description":"更难的真实软件工程任务；不同 harness 和任务版本不可直接合并。","owner":"SWE-bench","source":"https://arxiv.org/abs/2509.16941","sourceLabel":"SWE-bench Pro paper","sourceIds":["src-swebench-pro"],"lastVerified":"2026-08-28","versions":[{"id":"swebench-pro@v1","label":"v1","status":"active"}],"metrics":[{"id":"resolved","label":"Resolved","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":15,"profile":{"task":{"summary":"更接近真实软件维护的仓库级 issue 修复，强调长上下文、定位、实现和回归测试。","input":"issue、仓库快照与工程环境","output":"通过隐藏/公开测试的补丁"},"dataset":{"size":"官方 v1 任务集；规模、项目分布和子集以当前 release 为准","splits":["v1","project/difficulty slices"],"availability":"官方 leaderboard 与任务说明"},"protocol":{"mode":"system / agent harness","grader":"patch application + test-based resolution","tools":"repo access、terminal、网络、timeout 与 scaffold 需记录","sampling":"报告 solved instances、rollout 和失败分类"},"comparison":{"recommended":"固定 SWE-bench Pro version、harness、测试策略和上下文预算","avoid":"不要与 Verified/Lite 或不同 test-fix policy 的结果直接合并"},"source_locator":"SWE-bench Pro paper、官方 leaderboard 与 Scale 数据说明"}},{"id":"livecodebench","canonicalId":"livecodebench","short":"LCB","name":"LiveCodeBench","family":"coding","category":"coding","familyLabel":"代码","evaluationMode":"direct","comparisonKey":"livecodebench:livecodebench@v6:pass@1","metric":"pass@1","metricLabel":"Pass@1","scale":100,"unit":"%","direction":"higher","version":"V6","versionId":"livecodebench@v6","description":"持续收集、时间切分的竞赛编程评测；不同窗口不能混合。","owner":"LiveCodeBench","source":"https://livecodebench.github.io/leaderboard.html","sourceLabel":"LiveCodeBench official leaderboard","sourceIds":["livecodebench"],"lastVerified":"2026-08-28","versions":[{"id":"livecodebench@v5","label":"V5","status":"legacy"},{"id":"livecodebench@v6","label":"V6","status":"active"}],"metrics":[{"id":"pass@1","label":"Pass@1","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":16,"profile":{"task":{"summary":"持续收集新发布的竞赛编程题，覆盖代码生成、执行和测试输出预测。","input":"题目、约束和样例输入输出","output":"可执行程序或预测输出"},"dataset":{"size":"随题目发布日期和 cutoff 滚动；每个窗口题量不同","splits":["code generation","code execution","test output prediction","Easy/Medium/Hard"],"availability":"代码、题集版本与公开 leaderboard"},"protocol":{"mode":"model-only / direct code execution","grader":"沙箱执行与官方 tests，Pass@1 常用","tools":"语言、编译器、执行时限和是否可运行代码需披露","sampling":"start/end date、release version、temperature 和 contamination flag 必须记录"},"comparison":{"recommended":"同一时间窗口、release、语言和难度切片比较","avoid":"不要跨窗口混合；留意题目难度随时间变化和污染检测标记"},"source_locator":"LiveCodeBench official leaderboard 与 GitHub release/version README"}},{"id":"deepswe","canonicalId":"deepswe","short":"DeepSWE","name":"DeepSWE v1.1","family":"coding","category":"coding-agent","familyLabel":"代码 Agent","evaluationMode":"system","comparisonKey":"deepswe:deepswe@v1.1:resolved","metric":"resolved","metricLabel":"Resolved","scale":100,"unit":"%","direction":"higher","version":"v1.1","versionId":"deepswe@v1.1","description":"软件工程 Agent 任务；报告 harness、timeout 和 rollout 聚合方式。","owner":"DeepSWE","source":"https://deepswe.datacurve.ai/","sourceLabel":"DeepSWE leaderboard","sourceIds":["src-deepswe"],"lastVerified":"2026-08-28","versions":[{"id":"deepswe@v1.1","label":"v1.1","status":"active"}],"metrics":[{"id":"resolved","label":"Resolved","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":14,"profile":{"task":{"summary":"软件工程 Agent 在仓库环境中进行问题理解、修改、测试和修复闭环。","input":"issue/任务描述与代码仓库","output":"可验证的修复 patch"},"dataset":{"size":"DeepSWE v1.1：113 tasks","splits":["v1.1","task/project slices"],"availability":"official v1.1 leaderboard"},"protocol":{"mode":"system / agent harness","grader":"Resolved % + confidence interval；需保留 timeout 和失败类型","tools":"shell、编辑器、测试、网络和上下文预算是系统条件","sampling":"保留 config/effort、cost、output tokens、agent steps、rollout 聚合与成功定义"},"comparison":{"recommended":"同 harness、timeout、模型 endpoint 和任务版本内比较","avoid":"不要把一次成功 rollout 与多次搜索/自修复系统直接横比"},"source_locator":"deepswe.datacurve.ai official v1.1 leaderboard；113 tasks + CI/cost/output tokens/agent steps"}},{"id":"nl2repo","canonicalId":"nl2repo","short":"NL2Repo","name":"NL2Repo","family":"coding","category":"coding-agent","familyLabel":"代码 Agent","evaluationMode":"system","comparisonKey":"nl2repo:nl2repo@v1:resolved","metric":"resolved","metricLabel":"Resolved","scale":100,"unit":"%","direction":"higher","version":"v1","versionId":"nl2repo@v1","description":"从自然语言需求到仓库级实现；包含反作弊和判定设置。","owner":"NL2Repo","source":"https://github.com/multimodal-art-projection/NL2RepoBench","sourceLabel":"NL2Repo benchmark","sourceIds":["src-nl2repo"],"lastVerified":"2026-08-28","versions":[{"id":"nl2repo@v1","label":"v1","status":"active"}],"metrics":[{"id":"resolved","label":"Resolved","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"根据自然语言需求从 0 到 1 生成完整代码仓库，检验规划、跨文件实现与测试闭环。","input":"自然语言需求与隔离 runtime","output":"可运行的完整仓库"},"dataset":{"size":"104 tasks；packaged environment 中 103 runnable","splits":["task/repository slices","hidden original tests"],"availability":"official repositories、runtime 与评测说明公开"},"protocol":{"mode":"system / OpenHands-compatible agent harness","grader":"原始 pytest/hidden tests 的连续 passed-tests ratio","tools":"OpenHands/runtime version、CPU/RAM、repo access、依赖安装、网络和 timeout 需记录","sampling":"完整 repo、测试执行、reward 与失败原因要可追溯"},"comparison":{"recommended":"固定仓库 commit、scaffold、依赖环境和测试版本","avoid":"不要把仅生成代码的 pass@1 与仓库级 resolved 混排"},"source_locator":"multimodal-art-projection/NL2RepoBench + EnvCommons/NL2RepoBench official repositories；104 tasks / 103 runnable"}},{"id":"terminal-bench","canonicalId":"terminal-bench","short":"Terminal","name":"Terminal-Bench 2.1","family":"coding","category":"coding-agent","familyLabel":"代码 Agent","evaluationMode":"system","comparisonKey":"terminal-bench:terminal-bench@2.1:pass_rate","metric":"pass_rate","metricLabel":"Pass rate","scale":100,"unit":"%","direction":"higher","version":"2.1 · pinned comparison","versionId":"terminal-bench@2.1","description":"终端环境中的长程软件工程任务；执行器与 agent harness 是结果的一部分。","owner":"Terminal-Bench","source":"https://www.tbench.ai/","sourceLabel":"Terminal-Bench","sourceIds":["terminal-bench"],"lastVerified":"2026-08-28","versions":[{"id":"terminal-bench@2.0","label":"2.0","status":"legacy"},{"id":"terminal-bench@2.1","label":"2.1 · pinned comparison","status":"active"},{"id":"terminal-bench@3","label":"3 · shipped; separate track","status":"active"}],"metrics":[{"id":"pass_rate","label":"Pass rate","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":12,"profile":{"task":{"summary":"在隔离终端环境中完成长程工程任务，包含安装、调试、文件操作和验证。","input":"任务指令与预置终端/容器环境","output":"环境状态、文件产物或通过验证的解决方案"},"dataset":{"size":"2.0/2.1 是当前已收录结果窗口；Terminal-Bench 3 已发布并需作为独立 version/track","splits":["2.0","2.1 pinned","3 separate track","task/domain slices"],"availability":"官方任务、镜像和 leaderboard"},"protocol":{"mode":"system / terminal agent","grader":"任务级 verifier，常汇总为 pass rate","tools":"shell、网络、文件系统、容器镜像、timeout 和资源上限必须记录","sampling":"agent harness、模型 effort、重试和最终状态需保留"},"comparison":{"recommended":"固定 benchmark version、镜像、权限、timeout 和 harness","avoid":"不要把不同 terminal image 或人工介入的结果当作同一系统"},"source_locator":"tbench.ai + FrontierBench official pages；2.1 与 3 的任务、镜像、harness 和 verifier 不可 rolling overwrite"}},{"id":"bfcl","canonicalId":"bfcl","short":"BFCL","name":"Berkeley Function Calling Leaderboard","family":"agents","category":"tool-use","familyLabel":"工具调用","evaluationMode":"hybrid","comparisonKey":"bfcl@v4:accuracy:calling_mode:evaluator_commit","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"V4","versionId":"bfcl@v4","description":"函数调用、并行调用和多轮 Agent 任务；需区分 native FC 与 prompt workaround，并把 evaluator/tool protocol 视为系统条件。","owner":"Berkeley Gorilla Lab","source":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLabel":"Berkeley Function Calling Leaderboard","sourceIds":["src-bfcl"],"lastVerified":"2026-08-28","versions":[{"id":"bfcl@v4","label":"V4","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":17,"profile":{"task":{"summary":"覆盖单轮、多轮、并行、复杂和 agentic function calling 的工具调用能力。","input":"用户意图、工具 schema 与对话状态","output":"一个或多个结构化 function calls 及参数"},"dataset":{"size":"V4 按类别发布；子集数量随官方 CSV/commit 更新","splits":["simple","multiple","parallel","multi-turn","agentic","native_fc/prompt"],"availability":"官方 leaderboard CSV 与 evaluator code"},"protocol":{"mode":"hybrid / tool protocol","grader":"调用结构、参数和执行结果 evaluator；V4 overall 是子类别的未加权平均","tools":"native function calling 与 prompt workaround 必须分列","sampling":"保留 evaluator commit f7cf735、bfcl-eval package、schema、temperature 和错误类型"},"comparison":{"recommended":"同一 V4 commit、calling mode、模型 endpoint 和子类比较","avoid":"不要把 BFCL overall 当通用准确率，也不要合并 native FC 与 prompt 模拟"},"source_locator":"Berkeley Gorilla official leaderboard data_overall.csv 与 evaluator commit"}},{"id":"tau-bench","canonicalId":"tau-bench","short":"τ-bench","name":"τ-bench family","family":"agents","category":"tool-use","familyLabel":"工具调用","evaluationMode":"system","comparisonKey":"tau-bench:tau-bench@tau3-knowledge:pass^k","metric":"pass^k","metricLabel":"Pass^k","scale":100,"unit":"%","direction":"higher","version":"τ³ Knowledge","versionId":"tau-bench@tau3-knowledge","description":"用户—Agent—工具闭环；τ、τ²、τ³ Knowledge 与 τ³ Voice 是不同 release/track，必须按 domain、user simulator、Pass^k 和 voice protocol 分开。","owner":"τ-bench","source":"https://taubench.com/","sourceLabel":"τ-bench","sourceIds":["src-taubench"],"lastVerified":"2026-08-28","versions":[{"id":"tau-bench@v1","label":"τ-bench v1","status":"legacy"},{"id":"tau-bench@tau2","label":"τ²-bench","status":"active"},{"id":"tau-bench@tau3-knowledge","label":"τ³ Knowledge","status":"active"},{"id":"tau-bench@tau3-voice","label":"τ³ Voice","status":"active"}],"metrics":[{"id":"pass","label":"Pass","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"},{"id":"pass^k","label":"Pass^k","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":20,"profile":{"task":{"summary":"用户模拟器与 Agent 通过业务工具完成零售、航空等真实流程，检验策略遵循和状态变化。","input":"用户对话、业务规则、工具 schema 与数据库状态","output":"工具调用轨迹及最终业务状态"},"dataset":{"size":"按 τ、τ²、τ³ release 与领域变化；固定题量以版本快照为准","splits":["retail","airline","telecom","banking","τ²/τ³ variants"],"availability":"任务、模拟器和公开 leaderboard"},"protocol":{"mode":"system / user-agent-tool loop","grader":"数据库状态 verifier；Pass^k 衡量 k 次独立试验全成功概率","tools":"业务工具集合、用户 simulator、状态初始化和 turn budget 必须固定","sampling":"报告 pass@1/pass^k、领域、k、重试与对话轮数"},"comparison":{"recommended":"同一 release、领域、工具状态、user simulator 和 k 值比较","avoid":"不要把 Pass^k 当普通 accuracy，也不要跨 τ、τ²、τ³ 直接合并"},"source_locator":"τ-bench official site、domain/release说明与 Pass^k 定义"}},{"id":"toolathlon","canonicalId":"toolathlon","short":"Toolathlon","name":"Toolathlon","family":"agents","category":"tool-use","familyLabel":"工具调用","evaluationMode":"system","comparisonKey":"toolathlon:toolathlon@v1:success","metric":"success","metricLabel":"Success","scale":100,"unit":"%","direction":"higher","version":"verified","versionId":"toolathlon@v1","description":"跨工具、长程任务和执行可靠性；明确工具集合与预算。","owner":"Toolathlon","source":"https://toolathlon.ai/","sourceLabel":"Toolathlon benchmark","sourceIds":["src-toolathlon"],"lastVerified":"2026-08-28","versions":[{"id":"toolathlon@v1","label":"verified","status":"active"}],"metrics":[{"id":"success","label":"Success","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"跨真实软件应用的长程工具调用任务，要求 Agent 规划、探索、执行并达到可验证终态。","input":"自然语言用户请求、应用状态与 MCP/tool catalog","output":"跨应用工具轨迹和目标状态"},"dataset":{"size":"108 tasks、32 applications、604 tools（论文/官方介绍口径）","splits":["7 task categories","Toolathlon-Verified release"],"availability":"任务、ground truth、evaluator 与 trajectories 公开"},"protocol":{"mode":"system / multi-app tool agent","grader":"专用 deterministic state-based evaluation scripts","tools":"MCP server、应用初始状态、权限、网络和工具 schema 是测量对象","sampling":"记录 Pass@1、Pass@3、Pass^3、turns 和 cost"},"comparison":{"recommended":"固定 Verified release、工具版本、初始状态、模型与 agent scaffold","avoid":"不要把 tool-call accuracy 或单应用结果当整套任务 success"},"source_locator":"Toolathlon official introduction/docs 与 ICLR paper"}},{"id":"gdpval-aa","canonicalId":"gdpval-aa","short":"GDPval","name":"GDPval-AA","family":"agents","category":"knowledge-work","familyLabel":"知识工作","evaluationMode":"direct","comparisonKey":"gdpval-aa:gdpval-aa@v2:elo","metric":"elo","metricLabel":"Elo","scale":2500,"unit":"elo","direction":"higher","version":"AA","versionId":"gdpval-aa@v2","description":"知识工作任务的 pairwise/Elo 评价；Elo 不应转成准确率。","owner":"Artificial Analysis","source":"https://artificialanalysis.ai/methodology/intelligence-benchmarking","sourceLabel":"GDPval methodology","sourceIds":["src-gdpval"],"lastVerified":"2026-08-28","versions":[{"id":"gdpval-aa@v2","label":"AA","status":"active"}],"metrics":[{"id":"elo","label":"Elo","unit":"elo","scale":{"min":0,"max":2500},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"知识工作任务的成品由评审者两两比较，衡量相对质量、完成度和专业可用性。","input":"职业场景指令与必要材料","output":"报告、分析、计划或其他知识工作成品"},"dataset":{"size":"任务池与类别按 Artificial Analysis snapshot 变化","splits":["job/task categories","pairwise evaluation pool"],"availability":"聚合榜单；原始任务/投票通常不完全公开"},"protocol":{"mode":"preference / pairwise","grader":"人类或模型 pairwise judge，聚合为 Elo","tools":"材料、上下文、检索权限和评审模型需披露","sampling":"保留对手池、投票数、时间窗口和置信区间"},"comparison":{"recommended":"同一 snapshot、任务类别、评审池和 Elo 估计方法比较","avoid":"Elo 是相对偏好，不可转换成 accuracy 或跨榜单直接相加"},"source_locator":"Artificial Analysis GDPval methodology 与榜单说明"}},{"id":"mmmu","canonicalId":"mmmu","short":"MMMU","name":"MMMU","family":"multimodal","category":"multimodal","familyLabel":"多模态","evaluationMode":"direct","comparisonKey":"mmmu:mmmu@v1:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"original","versionId":"mmmu@v1","description":"大学级别的跨学科图文理解；原版与 Pro 版本必须分开记录。","owner":"MMMU","source":"https://mmmu-benchmark.github.io/","sourceLabel":"MMMU benchmark","sourceIds":["mmmu"],"lastVerified":"2026-08-28","versions":[{"id":"mmmu@v1","label":"original","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":null,"profile":{"task":{"summary":"大学级多学科图文理解与推理，题目来自考试、教材和专业图表。","input":"图像/图表与多项选择或短答案问题","output":"选项或短答案"},"dataset":{"size":"约 11.5K 多模态题，覆盖 6 个核心学科","splits":["validation","test","subject/discipline slices"],"availability":"公开数据与 leaderboard；test 答案按版本控制"},"protocol":{"mode":"model-only / multimodal direct","grader":"答案 exact-match accuracy","tools":"图像分辨率、OCR、外部工具和 image order 需记录","sampling":"zero/few-shot、选项顺序和答案抽取不可混用"},"comparison":{"recommended":"同一 original release、模态输入和 subject aggregation 比较","avoid":"不要与 MMMU-Pro、纯文本转写或工具增强结果直接平均"},"source_locator":"MMMU official site 与原始论文"}},{"id":"mmmu-pro","canonicalId":"mmmu-pro","short":"MMMU-Pro","name":"MMMU-Pro","family":"multimodal","category":"multimodal","familyLabel":"多模态","evaluationMode":"direct","comparisonKey":"mmmu-pro:mmmu-pro@v1:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"Pro","versionId":"mmmu-pro@v1","description":"MMMU 的更难版本：1,730 个 unique items，分别以 standard 与 vision-only presentation 评测；两种 setting 不能相加成 3,460 道独立题。","owner":"MMMU","source":"https://mmmu-benchmark.github.io/","sourceLabel":"MMMU benchmark","sourceIds":["mmmu"],"lastVerified":"2026-08-28","versions":[{"id":"mmmu-pro@v1","label":"Pro","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":true,"displayPriority":19,"profile":{"task":{"summary":"针对图像可忽略、文字捷径等问题设计的更难多模态选择题，要求真正读取视觉信息。","input":"图像/截图与十选一或 vision-only 问题","output":"选项答案"},"dataset":{"size":"1,730 个 unique items；每个 item 分别以 standard 与 vision-only presentation 评测","splits":["standard","vision-only","subject slices"],"availability":"官方 repository、paper、数据与 evaluator"},"protocol":{"mode":"multimodal model-only / direct","grader":"选项 exact-match accuracy","tools":"必须注明 standard/vision-only、图像输入、截图/原图格式和 OCR/工具","sampling":"direct/CoT、prompt、选项顺序和格式化规则需固定"},"comparison":{"recommended":"分别比较 standard 与 vision-only，并固定 image preprocessing","avoid":"不要把 MMMU-Pro 与原版 MMMU 或 text-only shortcut 结果混合"},"source_locator":"MMMU official repository mmmu-pro/evaluate.py；1,730 unique items 在 standard / vision-only 两种 setting 下分别评测"}},{"id":"mathvista","canonicalId":"mathvista","short":"MathVista","name":"MathVista","family":"multimodal","category":"multimodal","familyLabel":"多模态","evaluationMode":"direct","comparisonKey":"mathvista:mathvista@v1:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"v1","versionId":"mathvista@v1","description":"视觉数学推理；分辨率、OCR 和工具会影响结果。","owner":"MathVista","source":"https://mathvista.github.io/","sourceLabel":"MathVista · official project and dataset","sourceIds":["src-mathvista"],"lastVerified":"2026-08-28","versions":[{"id":"mathvista@v1","label":"v1","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"视觉数学推理，结合图表、几何、函数图像和自然语言解释。","input":"数学图像/图表与问题","output":"短答案、数值或选项"},"dataset":{"size":"6,141 examples，来自 31 个数据集；testmini 1,000、test 5,141","splits":["testmini","test","FQA/GPS/MWP/TQA/VQA slices"],"availability":"项目页、Hugging Face 数据与公开 leaderboard"},"protocol":{"mode":"multimodal model-only / direct","grader":"答案匹配 accuracy","tools":"分辨率、OCR、计算器、self-consistency 和图像预处理需记录","sampling":"testmini 适合开发；test 答案不公开，需按官方流程提交"},"comparison":{"recommended":"同一 test/testmini、图像输入和工具设置内比较","avoid":"不要把 testmini 分数与完整 test 或自制 OCR 文本结果直接横比"},"source_locator":"MathVista official site（testmini/test 划分）与原始论文"}},{"id":"video-mme","canonicalId":"video-mme","short":"Video-MME","name":"Video-MME","family":"multimodal","category":"multimodal","familyLabel":"多模态","evaluationMode":"direct","comparisonKey":"video-mme:video-mme@v1:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"v1","versionId":"video-mme@v1","description":"视频理解；需记录视频时长、音频和采样策略。","owner":"Video-MME","source":"https://video-mme.github.io/","sourceLabel":"Video-MME · official project","sourceIds":["src-video-mme"],"lastVerified":"2026-08-28","versions":[{"id":"video-mme@v1","label":"v1","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"视频理解问答，覆盖短/中/长视频、事件时序、字幕和音频信息。","input":"视频片段、问题与可选字幕/音频","output":"选项或短答案"},"dataset":{"size":"官方版本包含约 900 videos / 2,700 questions；具体 release 需注明","splits":["short","medium","long","subject slices"],"availability":"数据卡/论文与 leaderboard"},"protocol":{"mode":"multimodal model-only 或 video-agent","grader":"答案 accuracy","tools":"帧采样率、视频时长、音频/字幕开关、上下文预算需记录","sampling":"固定采样策略和答案抽取；工具增强另列"},"comparison":{"recommended":"同一 video release、duration bucket、audio/subtitle policy 比较","avoid":"不要把无音频、低帧率或摘要输入结果和完整视频结果混在一起"},"source_locator":"Video-MME project/paper 与数据集说明"}},{"id":"osworld","canonicalId":"osworld","short":"OSWorld","name":"OSWorld 2.0","family":"agents","category":"computer-use","familyLabel":"电脑操作","evaluationMode":"system","comparisonKey":"osworld:osworld-v2@2026.06.24:success","metric":"success","metricLabel":"Success","scale":100,"unit":"%","direction":"higher","version":"OSWorld-V2 · 2026.06.24 · 108 tasks","versionId":"osworld-v2@2026.06.24","description":"OSWorld-V2 的 108 个真实桌面任务；代码、tasks、assets、websites、provider images 与 release manifest 必须同版本。","owner":"OSWorld","source":"https://github.com/xlang-ai/OSWorld-V2","sourceLabel":"OSWorld-V2 benchmark and release manifest","sourceIds":["src-osworld"],"lastVerified":"2026-08-28","versions":[{"id":"osworld@2.0","label":"2.0 family label","status":"legacy"},{"id":"osworld-v2@2026.06.24","label":"OSWorld-V2 · 2026.06.24 · 108 tasks","status":"active"}],"metrics":[{"id":"success","label":"Success","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":18,"profile":{"task":{"summary":"在真实桌面和浏览器应用中完成多步操作，测视觉定位、规划、执行和状态验证。","input":"自然语言任务、桌面截图与应用状态","output":"最终 GUI 状态或生成的文件/表单"},"dataset":{"size":"OSWorld-V2 release osworld-v2-2026.06.24：108 tasks","splits":["desktop apps","browser","task/domain slices"],"availability":"official OSWorld-V2 code、tasks、assets、websites、provider images 与 release manifest"},"protocol":{"mode":"system / computer-use agent","grader":"环境状态检查与任务级 success","tools":"code/tasks/assets/websites/provider images、OS 镜像、分辨率、browser profile、截图频率和执行器必须同 release","sampling":"记录任务重试、timeout、人工介入和失败类别"},"comparison":{"recommended":"相同 OS image、app state、screen size、harness 和任务版本内比较","avoid":"不要将 GUI agent success 与文本指令遵循或网页静态代码分混排"},"source_locator":"xlang-ai/OSWorld-V2 + benchmark_releases/osworld-v2-2026.06.24.json"}},{"id":"ruler","canonicalId":"ruler","short":"RULER","name":"RULER","family":"retrieval","category":"long-context","familyLabel":"长上下文","evaluationMode":"direct","comparisonKey":"ruler:ruler@v2:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"v2","versionId":"ruler@v2","description":"长上下文检索与推理；按 context bucket 报告曲线而非单点。","owner":"RULER","source":"https://github.com/NVIDIA/RULER","sourceLabel":"RULER · official code and paper","sourceIds":["src-ruler"],"lastVerified":"2026-08-28","versions":[{"id":"ruler@v2","label":"v2","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"可调 context length 的合成检索与推理任务，超越单一 needle-in-a-haystack。","input":"含 distractors、needle、多跳关系或聚合要求的长上下文","output":"检索出的值、关系判断或聚合答案"},"dataset":{"size":"13 个任务，按 context bucket 动态生成","splits":["retrieval","multi-hop tracing","aggregation","question answering","4K–128K+ buckets"],"availability":"官方代码可生成任务；每次配置需保存"},"protocol":{"mode":"model-only / synthetic direct","grader":"程序化 exact-match accuracy","tools":"上下文长度、needle 数量、位置分布和 tokenization 是协议","sampling":"应报告每个 bucket 曲线，不只报总体平均"},"comparison":{"recommended":"固定生成 seed、task config、context buckets 和 max output","avoid":"不要用最大 advertised context 或单个 NIAH 分数替代 RULER 曲线"},"source_locator":"RULER paper 与 NVIDIA/RULER 官方代码"}},{"id":"longbench","canonicalId":"longbench","short":"LongBench","name":"LongBench v2","family":"retrieval","category":"long-context","familyLabel":"长上下文","evaluationMode":"direct","comparisonKey":"longbench:longbench@v2:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"v2","versionId":"longbench@v2","description":"多任务长文本理解；记录上下文长度、语言和检索策略。","owner":"LongBench","source":"https://longbench2.github.io/","sourceLabel":"LongBench v2 · official project","sourceIds":["src-longbench"],"lastVerified":"2026-08-28","versions":[{"id":"longbench@v2","label":"v2","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"真实长文本多任务理解与推理，覆盖单文档、多文档、长对话、代码仓库和结构化数据。","input":"8K–2M words 的长上下文与问题","output":"多项选择答案或任务特定输出"},"dataset":{"size":"LongBench v2 为 503 个 challenging multiple-choice questions","splits":["single-document QA","multi-document QA","long ICL","long dialogue","code repository","long structured data"],"availability":"项目页、数据与评测代码"},"protocol":{"mode":"model-only / long-context direct","grader":"选项 accuracy；可按六类切片","tools":"截断、检索、上下文拼接和语言设置需记录","sampling":"context length、语言、prompt 与答案抽取保持一致"},"comparison":{"recommended":"按六类和 context bucket 分别比较，同一 v2 release","avoid":"不要把 v1 与 v2、摘要/RAG 预处理和原始长上下文结果混合"},"source_locator":"LongBench v2 paper、project page 与数据卡"}},{"id":"ceval","canonicalId":"ceval","short":"C-Eval","name":"C-Eval","family":"multilingual","category":"chinese","familyLabel":"中文 / 多语言","evaluationMode":"direct","comparisonKey":"ceval:ceval@v2:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"v2","versionId":"ceval@v2","description":"中文多学科知识评测；与 CMMLU、SuperCLUE 分开看。","owner":"HKUST NLP","source":"https://github.com/hkust-nlp/ceval","sourceLabel":"C-Eval repository","sourceIds":["src-ceval"],"lastVerified":"2026-08-28","versions":[{"id":"ceval@v2","label":"v2","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"中文多学科选择题，涵盖基础教育、大学与专业知识，含高难度学科。","input":"中文题干、选项和学科标签","output":"选项答案"},"dataset":{"size":"13,948 题、52 个学科、4 个难度级别","splits":["dev","val","test","C-Eval Hard"],"availability":"官方 repo 与数据/评测脚本"},"protocol":{"mode":"model-only / direct","grader":"选项 exact-match accuracy，按学科聚合","tools":"shot、中文 prompt、选项顺序和答案解析需披露","sampling":"固定 v2 数据、subject weighting 和 few-shot 示例"},"comparison":{"recommended":"同一 v2 split、语言、shot 和学科聚合比较","avoid":"不要把 C-Eval Hard、全量均值与 CMMLU/SuperCLUE 直接合并"},"source_locator":"HKUST NLP C-Eval README（题量、学科、难度）与代码"}},{"id":"cmmlu","canonicalId":"cmmlu","short":"CMMLU","name":"CMMLU","family":"multilingual","category":"chinese","familyLabel":"中文 / 多语言","evaluationMode":"direct","comparisonKey":"cmmlu:cmmlu@v1:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"v1","versionId":"cmmlu@v1","description":"中文知识与推理评测；报告语言、shot 和 prompt。","owner":"CMMLU","source":"https://github.com/haonan-li/CMMLU","sourceLabel":"CMMLU repository","sourceIds":["src-cmmlu"],"lastVerified":"2026-08-28","versions":[{"id":"cmmlu@v1","label":"v1","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"中文知识与推理多任务选择题，覆盖生活常识、专业学科和社会科学。","input":"中文题干、选项与 subject 标签","output":"选项答案"},"dataset":{"size":"约 11.5K 题、67 个 subjects（以官方 release 为准）","splits":["dev","test","subject slices"],"availability":"官方 repository 与数据文件"},"protocol":{"mode":"model-only / direct","grader":"选项 exact-match accuracy","tools":"中文/英文 prompt、few-shot、选项顺序和答案解析需固定","sampling":"报告 overall 与 subject-level，不隐藏语言设置"},"comparison":{"recommended":"同一 v1 split、prompt 和 subject weighting 内比较","avoid":"不要将 CMMLU 与 C-Eval、翻译版 MMLU 或中文 Arena Elo 当同一指标"},"source_locator":"CMMLU official repository、paper 与数据说明"}},{"id":"superclue","canonicalId":"superclue","short":"SuperCLUE","name":"SuperCLUE Arena","family":"multilingual","category":"chinese","familyLabel":"中文 / 多语言","evaluationMode":"direct","comparisonKey":"superclue:superclue@arena:elo","metric":"elo","metricLabel":"Elo","scale":2500,"unit":"elo","direction":"higher","version":"Arena","versionId":"superclue@arena","description":"中文偏好竞技榜；属于 pairwise/Elo 评价，不等同于准确率。","owner":"SuperCLUE","source":"https://www.superclueai.com/arena","sourceLabel":"SuperCLUE arena","sourceIds":["src-superclue"],"lastVerified":"2026-08-28","versions":[{"id":"superclue@arena","label":"Arena","status":"active"}],"metrics":[{"id":"elo","label":"Elo","unit":"elo","scale":{"min":0,"max":2500},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"中文对话偏好竞技，用户或评审者对两个模型回答进行 pairwise 选择。","input":"中文用户请求与两个候选回答","output":"偏好投票"},"dataset":{"size":"随 Arena 时间窗口、category 和投票池动态变化","splits":["overall","category slices","latest snapshot"],"availability":"公开榜单与部分投票/快照数据"},"protocol":{"mode":"preference / pairwise","grader":"Bradley–Terry/Elo 聚合","tools":"endpoint、system prompt、评审池和过滤规则需记录","sampling":"保留 votes、置信区间、发布日期和榜单快照"},"comparison":{"recommended":"同一时间窗口、category、投票池和 rating 算法内比较","avoid":"Elo 不是 accuracy；不要跨快照或与英文 Arena 直接平均"},"source_locator":"SuperCLUE Arena 官方页面与榜单快照"}},{"id":"mmmlu","canonicalId":"mmmlu","short":"MMMLU","name":"MMMLU","family":"multilingual","category":"multilingual","familyLabel":"中文 / 多语言","evaluationMode":"direct","comparisonKey":"mmmlu:mmmlu@v1:accuracy","metric":"accuracy","metricLabel":"准确率","scale":100,"unit":"%","direction":"higher","version":"v1","versionId":"mmmlu@v1","description":"多语言 MMLU；按语言切片保留分数，避免平均值掩盖差异。","owner":"OpenAI","source":"https://huggingface.co/datasets/openai/MMMLU","sourceLabel":"MMMLU dataset","sourceIds":["src-mmmlu"],"lastVerified":"2026-08-28","versions":[{"id":"mmmlu@v1","label":"v1","status":"active"}],"metrics":[{"id":"accuracy","label":"准确率","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"专业人工翻译的多语言 MMLU 测试集，观察知识与推理在不同语言中的迁移。","input":"多语言选择题（含 57 类主题）","output":"选项答案"},"dataset":{"size":"MMLU test set 翻译为 14 种语言；主题与样本数按数据卡","splits":["14 locale configs","57 subject categories","test"],"availability":"OpenAI Hugging Face 数据集，MIT license"},"protocol":{"mode":"model-only / multilingual direct","grader":"选项 exact-match accuracy，按语言和主题切片","tools":"prompt 语言、翻译版本、选项顺序和答案解析需固定","sampling":"同时报告 macro language average 与 per-language scores"},"comparison":{"recommended":"固定 locale、subject weighting、prompt language 和 test snapshot","avoid":"不要只看总体平均掩盖低资源语言，也不要把 MMMLU 当新题集与 MMLU-Pro 混合"},"source_locator":"OpenAI MMMLU Hugging Face dataset card 与 Simple Evals references"}},{"id":"cybergym","canonicalId":"cybergym","short":"CyberGym","name":"CyberGym","family":"agents","category":"cyber","familyLabel":"网络安全","evaluationMode":"system","comparisonKey":"cybergym:cybergym@v1:success","metric":"success","metricLabel":"Success","scale":100,"unit":"%","direction":"higher","version":"v1","versionId":"cybergym@v1","description":"漏洞发现/利用能力；要区分发现、利用和工具权限。","owner":"CyberGym","source":"https://cybergym.ai/","sourceLabel":"CyberGym benchmark","sourceIds":["src-cybergym","src-cybergym-paper"],"lastVerified":"2026-08-28","versions":[{"id":"cybergym@v1","label":"v1","status":"active"}],"metrics":[{"id":"success","label":"Success","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"给定漏洞描述和未修复代码库，Agent 生成可执行 proof-of-concept 以复现真实漏洞。","input":"漏洞描述、源码/构建环境与服务接口","output":"可验证的 PoC 或 exploit trace"},"dataset":{"size":"1,507 个漏洞，来自 188 个开源项目（官方论文口径）","splits":["Level 1","project/language slices","vulnerability families"],"availability":"任务、服务器、镜像和 scorer 公开但资源规模大"},"protocol":{"mode":"system / cybersecurity agent","grader":"在隔离环境中运行 PoC，检查漏洞是否复现","tools":"编译器、Docker 网络、源码检索、shell 权限和 sandbox policy 必须记录","sampling":"发现、PoC 复现和修复不是同一指标；保留任务级轨迹"},"comparison":{"recommended":"固定 Level、task snapshot、环境镜像、网络和工具权限比较","avoid":"不要把安全研究者手工结果或 exploit discovery 与自动 PoC success 混为一谈"},"source_locator":"UC Berkeley CyberGym paper/repository 与 official benchmark page"}},{"id":"arena-text","canonicalId":"arena-text","short":"Arena Text","name":"Arena Text","family":"arena","category":"preference","familyLabel":"人类偏好","evaluationMode":"direct","comparisonKey":"arena-text:arena-text@latest:arena_score_bt","metric":"arena_score_bt","metricLabel":"Arena Score (Bradley–Terry)","scale":3000,"unit":"rating","direction":"higher","version":"latest snapshot","versionId":"arena-text@latest","description":"Text Arena 的 Bradley–Terry 人类偏好分；按 snapshot、category、style-control、置信区间和 vote count 保留口径。","owner":"Arena / LMArena","source":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceIds":["lmarena-hf-dataset"],"lastVerified":"2026-08-28","versions":[{"id":"arena-text@latest","label":"latest snapshot","status":"active"}],"metrics":[{"id":"arena_score_bt","label":"Arena Score (Bradley–Terry)","unit":"rating","scale":{"min":0,"max":3000},"direction":"higher"}],"featured":false,"displayPriority":1,"profile":{"task":{"summary":"文本对话 Arena 的真实用户 pairwise 偏好，比较两个 endpoint 的回答质量与体验。","input":"用户文本提示与两个模型回答","output":"用户偏好投票"},"dataset":{"size":"随 LMArena 投票和时间窗口动态变化","splits":["overall","category","latest snapshot"],"availability":"官方页面与 Hugging Face leaderboard dataset"},"protocol":{"mode":"preference / pairwise","grader":"Bradley–Terry Arena Score、95% 置信区间与 vote count","tools":"system prompt、endpoint、匿名化、投票过滤、category、style-control 与 methodology version 需记录","sampling":"保存 votes、snapshot date、model display name、style-control 状态与 CI"},"comparison":{"recommended":"同一 snapshot、category、投票池和 rating 版本内比较","avoid":"Arena rating 是相对偏好，不等于知识准确率或独立测试集分数"},"source_locator":"lmarena-ai/leaderboard-dataset official data card；rating methodology 自 2024 起为 Bradley–Terry，2025 起 text 默认 style-control"}},{"id":"arena-vision","canonicalId":"arena-vision","short":"Arena Vision","name":"Arena Vision","family":"arena","category":"preference","familyLabel":"人类偏好","evaluationMode":"direct","comparisonKey":"arena-vision:arena-vision@latest:arena_score_bt","metric":"arena_score_bt","metricLabel":"Arena Score (Bradley–Terry)","scale":3000,"unit":"rating","direction":"higher","version":"latest snapshot","versionId":"arena-vision@latest","description":"Vision Arena 的 Bradley–Terry 人类偏好分；与 Text Arena 分开，并保留 snapshot、category、style-control、置信区间和 votes。","owner":"Arena / LMArena","source":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceIds":["lmarena-hf-dataset"],"lastVerified":"2026-08-28","versions":[{"id":"arena-vision@latest","label":"latest snapshot","status":"active"}],"metrics":[{"id":"arena_score_bt","label":"Arena Score (Bradley–Terry)","unit":"rating","scale":{"min":0,"max":3000},"direction":"higher"}],"featured":false,"displayPriority":8,"profile":{"task":{"summary":"视觉/图文 Arena 的 pairwise 偏好，比较模型对图像理解、编辑或生成式请求的实际体验。","input":"图像与文本提示、两个候选回答","output":"偏好投票"},"dataset":{"size":"动态投票快照；图片类型和 category 随榜单更新","splits":["overall","vision categories","latest snapshot"],"availability":"Arena 页面与公开历史数据集"},"protocol":{"mode":"multimodal preference / pairwise","grader":"Bradley–Terry Arena Score、95% 置信区间与 vote count","tools":"图像分辨率、输入格式、生成 endpoint、过滤规则、style-control 与 methodology version 需保留","sampling":"记录 votes、category、style-control、时间窗口和置信区间"},"comparison":{"recommended":"同一 vision snapshot、输入模态和 category 内比较","avoid":"不要与 Text Arena、MMMU accuracy 或不同图像任务直接平均"},"source_locator":"lmarena-ai/leaderboard-dataset official data card；vision rating 需保留 snapshot 与 style-control 状态"}},{"id":"arena-webdev","canonicalId":"arena-webdev","short":"Arena WebDev","name":"Arena WebDev","family":"arena","category":"coding-agent","familyLabel":"代码 / 应用构建","evaluationMode":"system","comparisonKey":"arena-webdev:arena-webdev@latest:arena_score_bt","metric":"arena_score_bt","metricLabel":"Arena Score (Bradley–Terry)","scale":3000,"unit":"rating","direction":"higher","version":"latest snapshot","versionId":"arena-webdev@latest","description":"Code/WebDev Arena rating；模型名可能包含 endpoint 或 effort tier。","owner":"Arena / LMArena","source":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceIds":["lmarena-hf-dataset"],"lastVerified":"2026-08-28","versions":[{"id":"arena-webdev@latest","label":"latest snapshot","status":"active"}],"metrics":[{"id":"arena_score_bt","label":"Arena Score (Bradley–Terry)","unit":"rating","scale":{"min":0,"max":3000},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"用户比较模型生成网页或应用的视觉质量、交互完成度和可用性。","input":"Web/app 构建需求与运行环境","output":"可运行网页/应用及用户偏好投票"},"dataset":{"size":"随 WebDev Arena prompt、投票与 snapshot 动态变化","splits":["webdev categories","latest snapshot"],"availability":"Arena leaderboard/data snapshot"},"protocol":{"mode":"system / coding-agent preference","grader":"用户 pairwise preference 聚合为 Bradley–Terry Arena Score","tools":"browser、runtime、scaffold、endpoint、effort tier 和模型名需记录","sampling":"保存 prompt、render environment、votes、置信区间和时间窗口"},"comparison":{"recommended":"固定 scaffold、浏览器/runtime、prompt category、snapshot 和 endpoint 比较","avoid":"不能把 WebDev preference score 当通用代码 pass@1，也不要忽略渲染环境"},"source_locator":"Arena WebDev leaderboard snapshot 与官方数据集"}},{"id":"arena-search","canonicalId":"arena-search","short":"Arena Search","name":"Arena Search","family":"arena","category":"tool-use","familyLabel":"搜索 / 工具","evaluationMode":"system","comparisonKey":"arena-search:arena-search@latest:arena_score_bt","metric":"arena_score_bt","metricLabel":"Arena Score (Bradley–Terry)","scale":3000,"unit":"rating","direction":"higher","version":"latest snapshot","versionId":"arena-search@latest","description":"Search Arena rating；与无搜索的 Text Arena 不直接合并。","owner":"Arena / LMArena","source":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceIds":["lmarena-hf-dataset"],"lastVerified":"2026-08-28","versions":[{"id":"arena-search@latest","label":"latest snapshot","status":"active"}],"metrics":[{"id":"arena_score_bt","label":"Arena Score (Bradley–Terry)","unit":"rating","scale":{"min":0,"max":3000},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"带搜索工具的 Arena，比较模型检索、引用、事实核验与综合回答体验。","input":"用户问题、搜索环境与两个候选系统","output":"带引用的回答与用户偏好投票"},"dataset":{"size":"随 Search Arena category、搜索引擎和 snapshot 动态变化","splits":["search category","latest snapshot"],"availability":"Arena page/historical dataset（字段随版本变化）"},"protocol":{"mode":"system / search-agent preference","grader":"pairwise user vote → Bradley–Terry Arena Score","tools":"搜索引擎、调用次数、浏览预算、时效与 citation policy 必须固定","sampling":"记录 query time、source freshness、votes、CI 与 snapshot"},"comparison":{"recommended":"同一搜索 provider、工具预算、category 和时间窗口比较","avoid":"不要与无搜索 Text Arena 或静态 accuracy 混合"},"source_locator":"Arena Search leaderboard/data snapshot 与 LMArena 说明"}},{"id":"arena-document","canonicalId":"arena-document","short":"Arena Document","name":"Arena Document","family":"arena","category":"multimodal","familyLabel":"文档理解","evaluationMode":"direct","comparisonKey":"arena-document:arena-document@latest:arena_score_bt","metric":"arena_score_bt","metricLabel":"Arena Score (Bradley–Terry)","scale":3000,"unit":"rating","direction":"higher","version":"latest snapshot","versionId":"arena-document@latest","description":"Document Arena rating；保留 category、votes 和置信区间。","owner":"Arena / LMArena","source":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceIds":["lmarena-hf-dataset"],"lastVerified":"2026-08-28","versions":[{"id":"arena-document@latest","label":"latest snapshot","status":"active"}],"metrics":[{"id":"arena_score_bt","label":"Arena Score (Bradley–Terry)","unit":"rating","scale":{"min":0,"max":3000},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"文档理解 Arena，比较模型处理长文档、表格、版面和跨页信息的用户偏好。","input":"文档/PDF/表格与文本问题","output":"回答、提取或总结及偏好投票"},"dataset":{"size":"动态文档 prompt 与投票快照","splits":["document categories","latest snapshot"],"availability":"Arena leaderboard/data snapshot"},"protocol":{"mode":"multimodal/document system preference","grader":"pairwise preference → rating","tools":"PDF parser、OCR、上下文长度、视觉输入和 endpoint 需记录","sampling":"保存 votes、category、文档类型和置信区间"},"comparison":{"recommended":"固定 parser、文档类型、上下文预算与 snapshot 比较","avoid":"不要把文档 Arena rating 与 MMMU/MathVista accuracy 直接合并"},"source_locator":"Arena Document leaderboard 与公开历史数据集"}},{"id":"arena-agent","canonicalId":"arena-agent","short":"Agent Arena","name":"Agent Arena","family":"agents","category":"coding-agent","familyLabel":"Agent 系统","evaluationMode":"system","comparisonKey":"arena-agent:arena-agent@latest:ips","metric":"ips","metricLabel":"IPS score","scale":{"min":-1,"max":1},"unit":"fraction","direction":"higher","version":"latest snapshot","versionId":"arena-agent@latest","description":"Agent Arena 的 IPS (τ̂) 分数；不是 Bradley–Terry Elo，单独作为 system run。","owner":"Arena / LMArena","source":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceIds":["lmarena-hf-dataset"],"lastVerified":"2026-08-28","versions":[{"id":"arena-agent@latest","label":"latest snapshot","status":"active"}],"metrics":[{"id":"ips","label":"IPS score","unit":"fraction","scale":{"min":-1,"max":1},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"多步 Agent 任务的系统级比较，关注完成、效率和轨迹质量，而非单轮回答。","input":"任务目标、工具/环境和 agent scaffold","output":"任务终态与完整 action trace"},"dataset":{"size":"随 Agent Arena task pool 和 snapshot 更新","splits":["task categories","latest snapshot"],"availability":"Arena 公开 leaderboard/data 字段可能变化"},"protocol":{"mode":"system / agent evaluation","grader":"IPS (τ̂) 等系统指标；不是 Bradley–Terry Elo","tools":"harness、工具、环境、预算、重试和任务池必须记录","sampling":"保留成功率、轨迹长度、成本与 IPS 估计细节"},"comparison":{"recommended":"同一 task pool、harness、budget 和 IPS 版本内比较","avoid":"不要将 IPS 当 Elo 或 accuracy，也不要剥离 scaffold 冒充模型-only 分数"},"source_locator":"Arena Agent leaderboard methodology 与数据集字段说明"}},{"id":"agents-last-exam","canonicalId":"agents-last-exam","short":"ALE","name":"Agents' Last Exam","family":"agents","category":"agent-workflow","familyLabel":"Agent 工作流","evaluationMode":"system","comparisonKey":"agents-last-exam:agents-last-exam@v1:pass_rate","metric":"pass_rate","metricLabel":"Pass rate","scale":100,"unit":"%","direction":"higher","version":"ALE-V1 (public leaderboard)","versionId":"agents-last-exam@v1","description":"ALE-V1 的长程、可验证专业工作流；按 full、near-term、full-spectrum、last-exam、licensed/unlicensed 和 ALE-CLI tracks 保留 split，不把不同 harness 混成裸模型分数。","owner":"UC Berkeley RDI","source":"https://agents-last-exam.org/leaderboard","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceIds":["agents-last-exam"],"lastVerified":"2026-08-28","versions":[{"id":"agents-last-exam@v1","label":"ALE-V1 (public leaderboard)","status":"active"}],"metrics":[{"id":"pass_rate","label":"Pass rate","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"},{"id":"avg_score","label":"Average score","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":13,"profile":{"task":{"summary":"真实职业工作流的长程 Agent 系统评测，任务需要工具、环境状态和多步交付。","input":"职业任务指令、受控环境与工具","output":"可审计的任务交付物/环境状态"},"dataset":{"size":"ALE-V1 snapshot：147 个公开任务、55 个子领域；living framework 约 150 public tasks，来自 1,000+ task corpus","splits":["full/overall","near-term","full-spectrum","last-exam","licensed","unlicensed","ALE-CLI/Linux"],"availability":"官方 leaderboard/API；snapshot task_count 与 living corpus size 分开记录"},"protocol":{"mode":"system / Agent × harness × environment × task","grader":"deterministic 与 judge graders；Pass Rate + partial-credit Score","tools":"harness、环境、effort、工具、成本、runtime 和 token budget 需记录","sampling":"保留 split、variant、harness、pass_rate、avg_score 与 observed snapshot"},"comparison":{"recommended":"只比较相同 split、harness、environment、effort 和 ALE version","avoid":"不要把 ALE 当裸模型分数，也不要与 SakanaAI ALE-Bench/Atari ALE 混名"},"source_locator":"agents-last-exam.org docs、official leaderboard 与公开 JSON endpoint"}},{"id":"ale-bench","canonicalId":"ale-bench","short":"ALE-Bench","name":"ALE-Bench","family":"coding","category":"algorithm-engineering","familyLabel":"算法工程","evaluationMode":"system","comparisonKey":"ale-bench:ale-bench@v1.2.1:performance","metric":"performance","metricLabel":"Performance score","scale":{"min":-1000000,"max":10000000},"unit":"score","direction":"higher","version":"v1.2.1","versionId":"ale-bench@v1.2.1","description":"基于 AtCoder Heuristic Contest 任务的长程、目标驱动算法工程；连续 performance/rank 不能与 pass/fail coding benchmark 直接合并。当前仅登记官方 leaderboard artifact，未启用自动抓取。","owner":"SakanaAI","source":"https://sakanaai.github.io/ALE-Bench-Leaderboard/","sourceLabel":"SakanaAI · ALE-Bench leaderboard artifact (catalog only)","sourceIds":["src-ale-bench"],"lastVerified":"2026-08-28","versions":[{"id":"ale-bench@v1.2.1","label":"v1.2.1","status":"active"}],"metrics":[{"id":"performance","label":"Performance score","unit":"score","scale":{"min":-1000000,"max":10000000},"direction":"higher"},{"id":"rank","label":"Contest rank","unit":"rank","scale":{"min":1,"max":100000},"direction":"lower"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"AtCoder Heuristic Contest 上的长程算法工程：Agent 迭代修改求解器并优化无已知精确解问题。","input":"优化问题、评分器、代码仓库与运行约束","output":"可执行 solver、绝对 performance score 与 rank"},"dataset":{"size":"问题族与比赛实例随 ALE-Bench release；官方 leaderboard artifact 为主","splits":["all","short","long","lite","self-refine"],"availability":"官方 repo/leaderboard；private evaluation 约束需保留"},"protocol":{"mode":"system / algorithm-engineering agent","grader":"contest scorer，按 performance/rank 评估，不是 pass/fail","tools":"固定 AWS/硬件、运行时间、编译器、迭代预算和 judge version","sampling":"保存绝对分、相对基线、rank、hardware 和 lite/full track"},"comparison":{"recommended":"固定 problem family、hardware、budget、judge 与 track 比较；performance 只在同一问题/基线口径内解读","avoid":"raw performance 的尺度随问题与评分器变化，不要跨 problem family 把绝对分数排成总榜，也不要与 Agents’ Last Exam、SWE resolved 或普通 pass@1 合并"},"source_locator":"SakanaAI ALE-Bench repository 与 leaderboard artifact；独立于 ALE-V1"}},{"id":"browsecomp","canonicalId":"browsecomp","short":"BrowseComp","name":"BrowseComp","family":"agents","category":"browsing-agent","familyLabel":"浏览 Agent","evaluationMode":"system","comparisonKey":"browsecomp:browsecomp@v1:accuracy","metric":"accuracy","metricLabel":"Accuracy","scale":100,"unit":"%","direction":"higher","version":"v1","versionId":"browsecomp@v1","description":"需要多步网页检索与证据组合的 browsing benchmark；分数依赖 agent、搜索工具、effort 与评测窗口。官方页面目前没有稳定 leaderboard API。","owner":"OpenAI","source":"https://openai.com/index/browsecomp/","sourceLabel":"OpenAI · BrowseComp benchmark","sourceIds":["src-browsecomp"],"lastVerified":"2026-08-28","versions":[{"id":"browsecomp@v1","label":"v1","status":"active"}],"metrics":[{"id":"accuracy","label":"Accuracy","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"多步网页检索与证据拼接，要求 browsing Agent 找到分散事实并给出可验证答案。","input":"复杂事实问题、开放网页和搜索工具","output":"最终答案及支持证据"},"dataset":{"size":"1,266 个 browsing problems（官方介绍口径）","splits":["problem categories","retrieval/evidence slices"],"availability":"官方任务说明公开；稳定 leaderboard/API 不承诺"},"protocol":{"mode":"system / browsing agent","grader":"答案 accuracy 与证据/验证规则","tools":"搜索 provider、浏览预算、页面快照、citation policy 和 effort 必须记录","sampling":"保留检索日期、网页状态、工具调用数和答案验证器"},"comparison":{"recommended":"同一网页快照、搜索工具、预算和评测窗口比较","avoid":"不要把 no-browse QA 或 provider 自报值当成相同协议"},"source_locator":"OpenAI BrowseComp official page、任务/方法说明"}},{"id":"aider-polyglot","canonicalId":"aider-polyglot","short":"Aider Polyglot","name":"Aider Polyglot","family":"coding","category":"coding","familyLabel":"代码","evaluationMode":"system","comparisonKey":"aider-polyglot:aider-polyglot@public:pass_rate","metric":"pass_rate","metricLabel":"Pass rate","scale":100,"unit":"%","direction":"higher","version":"public leaderboard","versionId":"aider-polyglot@public","description":"多语言代码编辑任务；保留 edit format、model endpoint、repo/task snapshot 与 pass rate。官方榜单为动态网页，未假定存在稳定 API。","owner":"Aider","source":"https://aider.chat/docs/leaderboards/","sourceLabel":"Aider · Polyglot coding leaderboard","sourceIds":["src-aider-polyglot"],"lastVerified":"2026-08-28","versions":[{"id":"aider-polyglot@public","label":"public leaderboard","status":"active"}],"metrics":[{"id":"pass_rate","label":"Pass rate","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"真实编辑器工作流中的多语言代码修改，要求模型正确应用 patch 并通过项目测试。","input":"仓库文件、自然语言 edit request 与测试","output":"符合 edit format 的代码修改"},"dataset":{"size":"225 exercises，覆盖 C++、Go、Java、JavaScript、Python、Rust","splits":["6 language slices","public leaderboard snapshot"],"availability":"official leaderboard 与 task repository"},"protocol":{"mode":"system / Aider code-edit harness","grader":"Percent correct（相对 attempts）以及 pass_rate_1 / pass_rate_2 分列","tools":"Aider version/commit、edit format、repo snapshot、endpoint、temperature 和 patch application 需记录","sampling":"保留日期、cost、commit、语言、attempt count 和失败类型"},"comparison":{"recommended":"固定 Aider commit、edit format、repo/task snapshot 和解码设置","avoid":"不要将 Aider 编辑成功率与纯代码生成 pass@1 或 SWE-bench resolved 直接合并"},"source_locator":"aider.chat official leaderboards + Polyglot benchmark release；225 exercises / 6 languages"}},{"id":"androidworld","canonicalId":"androidworld","short":"AndroidWorld","name":"AndroidWorld","family":"agents","category":"computer-use","familyLabel":"电脑操作","evaluationMode":"system","comparisonKey":"androidworld:androidworld@public:success","metric":"success","metricLabel":"Success","scale":100,"unit":"%","direction":"higher","version":"public","versionId":"androidworld@public","description":"Android emulator 中的真实应用操作任务；环境镜像、设备配置、harness 与任务版本决定可比性。","owner":"Google Research","source":"https://github.com/google-research/android_world","sourceLabel":"Google Research · AndroidWorld","sourceIds":["src-androidworld"],"lastVerified":"2026-08-28","versions":[{"id":"androidworld@public","label":"public","status":"active"}],"metrics":[{"id":"success","label":"Success","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"在 Android emulator 的真实应用中完成点击、输入、导航、编辑和状态查询任务。","input":"自然语言任务、Android app 状态和设备屏幕","output":"设备系统状态达到目标"},"dataset":{"size":"116 个 programmatic tasks、20 个真实 Android apps；支持动态参数化变体","splits":["app/task categories","dynamic instantiations"],"availability":"代码、emulator setup、任务初始化和 success checks 公开"},"protocol":{"mode":"system / mobile computer-use agent","grader":"基于设备状态的 durable reward/success checker","tools":"Android image、屏幕尺寸、应用版本、ADB/executor、网络和 reset 必须记录","sampling":"任务变体会影响结果；记录 seed、retry、timeout 和环境版本"},"comparison":{"recommended":"固定 emulator/image、app version、task seed、screen size 和 harness","avoid":"不要将动态变体的一次运行与固定桌面 OS benchmark 直接横比"},"source_locator":"Google Research AndroidWorld official site、paper 与 repository"}},{"id":"mle-bench","canonicalId":"mle-bench","short":"MLE-bench","name":"MLE-bench","family":"coding","category":"ml-engineering","familyLabel":"ML 工程","evaluationMode":"system","comparisonKey":"mle-bench:mle-bench@public:success","metric":"success","metricLabel":"Success","scale":100,"unit":"%","direction":"higher","version":"public","versionId":"mle-bench@public","description":"Kaggle 机器学习工程任务；按 Lite/Medium/High/All 与 grading report 分轨，不能与普通代码 pass@1 合并。","owner":"OpenAI","source":"https://github.com/openai/mle-bench/","sourceLabel":"OpenAI · MLE-bench","sourceIds":["src-mle-bench"],"lastVerified":"2026-08-28","versions":[{"id":"mle-bench@public","label":"public","status":"active"}],"metrics":[{"id":"success","label":"Success","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"把 Kaggle 机器学习竞赛改造成 Agent 工程任务，覆盖数据处理、训练、实验和提交。","input":"竞赛说明、数据、运行环境与本地 grading 工具","output":"模型/预测文件、提交结果和可复现实验"},"dataset":{"size":"75 个 Kaggle competitions、15 个问题类别","splits":["Lite/Low","Medium","High","All","competition categories"],"availability":"代码、离线数据与 grading logic 公开；新 leaderboard 提交可能暂停"},"protocol":{"mode":"system / ML-engineering agent","grader":"competition metric 与 grading report；常按达到 Kaggle medal baseline 聚合","tools":"GPU/CPU、时间/存储、数据访问、网络和 agent scaffold 必须记录","sampling":"报告每个 competition、复杂度 split、运行时和重复试验区间"},"comparison":{"recommended":"固定 competition snapshot、资源预算、agent、LLM 和 grading report 版本","avoid":"不要将 Kaggle medal-style success 与普通代码 pass@1 或单一平均分混合"},"source_locator":"OpenAI MLE-bench repository README、论文与 leaderboard table"}},{"id":"livebench","canonicalId":"livebench","short":"LiveBench","name":"LiveBench","family":"reasoning","category":"reasoning","familyLabel":"滚动综合评测","evaluationMode":"direct","comparisonKey":"livebench:livebench@rolling:Global average","metric":"Global average","metricLabel":"Global average","scale":100,"unit":"%","direction":"higher","version":"dated rolling releases","versionId":"livebench@rolling","description":"按日期滚动发布的综合评测；题目、类别、judge 与模型 endpoint 必须随 release 一起记录。","owner":"LiveBench","source":"https://livebench.ai/","sourceLabel":"LiveBench","sourceIds":["src-livebench","src-epoch-benchmark-hub"],"lastVerified":"2026-08-28","versions":[{"id":"livebench@rolling","label":"dated rolling releases","status":"active"}],"metrics":[{"id":"Global average","label":"Global average","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":4,"profile":{"task":{"summary":"按日期滚动加入新题的综合直接评测，覆盖数学、代码、推理、语言、指令遵循与数据分析等类别，以降低静态题库污染的影响。","input":"指定 release 中各 category/subtask 的文本或代码题","output":"答案、代码或结构化响应，由对应任务 grader 判定"},"dataset":{"size":"题量随 dated release 与 subtask 更新，不登记固定总题量","splits":["dated release","category","subtask（当前公开 evidence 还包含 23 个 subtask slices）"],"availability":"LiveBench 官方 release/leaderboard 持续更新；本目录已有一层 Epoch 外部聚合快照"},"protocol":{"mode":"model-only / direct","grader":"按 subtask 使用可执行、规则或 judge grader，再汇总 Global average","tools":"模型 endpoint、prompt、judge、工具许可和 category 必须随 release 记录","sampling":"固定 release date、解码设置与每题采样预算；总分不能替代 category/subtask 分数"},"comparison":{"recommended":"同一 dated release、category/subtask、judge、endpoint 和采样预算内比较","avoid":"不要跨 release 混排，也不要把 Epoch 聚合行的模型发布日期当作 LiveBench 数据版本"},"source_locator":"LiveBench 官方 dated release/leaderboard（src-livebench）；当前聚合 evidence 定位到 Epoch benchmark_data.zip 的 live_bench_external.csv:row=*;column=Global average"}},{"id":"epoch-arc_agi_2_external","canonicalId":"epoch-arc_agi_2_external","short":"ARC-AGI-2","name":"ARC-AGI-2 (Epoch external aggregation)","family":"reasoning","category":"reasoning","familyLabel":"抽象推理","evaluationMode":"unknown","comparisonKey":"epoch-arc_agi_2_external:epoch-arc_agi_2_external@rolling:Score","metric":"Score","metricLabel":"Score","scale":1,"unit":"fraction","direction":"higher","version":"Epoch external aggregation (rolling snapshot)","versionId":"epoch-arc_agi_2_external@rolling","description":"ARC-AGI-2 外部公开结果的 Epoch 聚合切片；原始 harness、预算与私有评测条件需回到逐行来源核对。","owner":"ARC Prize / Epoch AI aggregation","source":"https://epoch.ai/benchmarks/use-this-data","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceIds":["src-epoch-benchmark-hub"],"lastVerified":"2026-08-28","versions":[{"id":"epoch-arc_agi_2_external@rolling","label":"Epoch external aggregation (rolling snapshot)","status":"active"}],"metrics":[{"id":"Score","label":"Score","unit":"fraction","scale":{"min":0,"max":1},"direction":"higher"}],"featured":false,"displayPriority":6,"profile":{"task":{"summary":"ARC-AGI-2 要求从少量输入/输出彩色网格示例归纳抽象转换规则，并把规则应用到新的测试网格。","input":"训练示例网格对与待转换测试网格","output":"预测的目标网格"},"dataset":{"size":"公开/私有 evaluation 内容和规模按 ARC-AGI-2 release；Epoch 表为外部公开结果聚合，不假定固定题量","splits":["ARC-AGI-2 evaluation track","external reported rows"],"availability":"任务定义由 ARC Prize 发布；本条目录记录 Epoch external aggregation"},"protocol":{"mode":"unknown / external aggregation","grader":"源表报告 Score（fraction）；具体官方或第三方 grid exact-match 口径需追踪原始来源","tools":"搜索、代码、迭代、token/compute budget 与 private evaluation 条件在 Epoch 聚合列中不完整","sampling":"同一模型可能按 effort/setting 出现多行，必须保留原行 locator 与 setting"},"comparison":{"recommended":"仅在确认同一 ARC-AGI-2 track、评测集、工具与计算预算后作条件比较","avoid":"不要与 ARC-AGI-1 合并，不要把 Epoch external aggregation 当成统一复现实验或官方单一 harness"},"source_locator":"Epoch AI Benchmarking Hub downloadable snapshot（src-epoch-benchmark-hub）：benchmark_data.zip / arc_agi_2_external.csv:row=*;column=Score；逐行原始 source_link/notes 仍需核对"}},{"id":"epoch-frontiermath_tier_4","canonicalId":"epoch-frontiermath_tier_4","short":"FrontierMath T4 v2","name":"FrontierMath Tier 4 v2 (reported snapshot)","family":"reasoning","category":"math","familyLabel":"数学","evaluationMode":"system","comparisonKey":"epoch-frontiermath_tier_4:frontiermath@v2-tier-4:mean_score","metric":"mean_score","metricLabel":"Mean score","scale":100,"unit":"%","direction":"higher","version":"v2 · Tier 4 · 43 problems","versionId":"frontiermath@v2-tier-4","description":"FrontierMath v2 Tier 4 的 43 题 tool-enabled 切片；保留 Python tool、1M-token hard limit、answer() grader 与 Epoch snapshot locator。","owner":"Epoch AI","source":"https://epoch.ai/benchmarks/frontiermath-tier-4-v2","sourceLabel":"Epoch AI · FrontierMath Tier 4 v2","sourceIds":["src-frontiermath-v2","src-epoch-benchmark-hub"],"lastVerified":"2026-08-28","versions":[{"id":"epoch-frontiermath_tier_4@rolling","label":"pre-v2 rolling snapshot","status":"legacy"},{"id":"frontiermath@v2-tier-4","label":"v2 · Tier 4 · 43 problems","status":"active"}],"metrics":[{"id":"mean_score","label":"Mean score","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":7,"profile":{"task":{"summary":"FrontierMath Tier 4 的高难度研究数学问题，要求给出可判定的答案或推导，难度切片独立于 T1–T3。","input":"Tier 4 数学问题与必要定义","output":"可由官方/专家 grader 判定的答案或证明"},"dataset":{"size":"FrontierMath v2 Tier 4：43 题；v2 于 2026-06-12 发布且与 pre-v2 snapshot 不可合并","splits":["v2 Tier 4","Epoch reported snapshot"],"availability":"题面和答案可能受限；公开表提供带 locator 的聚合结果"},"protocol":{"mode":"system / tool-enabled source-reported results","grader":"官方 answer() grader / source mean_score；保留 grader 与 partial-credit 口径","tools":"Python tool、1M-token hard limit、CAS、检索、proof checker、reasoning effort 与采样预算必须记录","sampling":"保留 v2、model setting、题目窗口、judge 和采样次数；模型发布日期不是 benchmark version"},"comparison":{"recommended":"同一 Tier 4 release/window、工具、judge 与采样预算内比较","avoid":"不要与 FrontierMath T1–T3 直接平均，也不要跨保密窗口或不同 grader 排名"},"source_locator":"Epoch AI FrontierMath Tier 4 v2 official page + Benchmarking Hub snapshot：benchmark_data.zip / frontiermath_tier_4.csv:row=*;column=mean_score"}},{"id":"epoch-bbh_external","canonicalId":"epoch-bbh_external","short":"BBH","name":"BIG-Bench Hard (Epoch external aggregation)","family":"reasoning","category":"reasoning","familyLabel":"推理","evaluationMode":"unknown","comparisonKey":"epoch-bbh_external:epoch-bbh_external@rolling:Average","metric":"Average","metricLabel":"Average","scale":1,"unit":"fraction","direction":"higher","version":"Epoch external aggregation (rolling snapshot)","versionId":"epoch-bbh_external@rolling","description":"BIG-Bench Hard 外部结果的 Epoch 聚合切片；CoT、few-shot、任务平均与答案解析条件可能不同。","owner":"Google / Epoch AI aggregation","source":"https://epoch.ai/benchmarks/use-this-data","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceIds":["src-epoch-benchmark-hub"],"lastVerified":"2026-08-28","versions":[{"id":"epoch-bbh_external@rolling","label":"Epoch external aggregation (rolling snapshot)","status":"active"}],"metrics":[{"id":"Average","label":"Average","unit":"fraction","scale":{"min":0,"max":1},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"BIG-Bench Hard（BBH）汇集 23 个对早期语言模型较难的推理任务，覆盖逻辑、算法、常识和符号操作。","input":"各 BBH task 的自然语言问题与按协议给出的 demonstrations","output":"短答案、标签或选择项"},"dataset":{"size":"23 个 task；每个 task 的样例选择与评测规模按具体 release/protocol","splits":["23 task slices","external reported aggregate"],"availability":"BBH 数据公开；这里登记 Epoch 聚合的外部公开行"},"protocol":{"mode":"unknown / external aggregation","grader":"Epoch 源表报告跨任务 Average（fraction）；任务权重与答案解析需追踪原始来源","tools":"CoT、few-shot、prompt、解码和自洽采样可能因来源而异","sampling":"必须保留 shot/CoT、task aggregation、model setting 与原始行 locator"},"comparison":{"recommended":"确认相同 23-task release、shot/CoT、任务加权和答案解析后再比较","avoid":"不要把不同论文/厂商的外部行视为统一 harness；旧题污染和提示差异不可忽略"},"source_locator":"Epoch AI Benchmarking Hub downloadable snapshot（src-epoch-benchmark-hub）：benchmark_data.zip / bbh_external.csv:row=*;column=Average；原始 source_link/notes 决定协议解释"}},{"id":"helm-ifeval","canonicalId":"helm-ifeval","short":"HELM IFEval","name":"IFEval · HELM Capabilities","family":"reasoning","category":"reliability","familyLabel":"指令遵循","evaluationMode":"direct","comparisonKey":"helm-ifeval:helm-ifeval@v1.15.0:ifeval-strict-acc","metric":"ifeval-strict-acc","metricLabel":"IFEval strict accuracy","scale":1,"unit":"fraction","direction":"higher","version":"HELM Capabilities v1.15.0","versionId":"helm-ifeval@v1.15.0","description":"HELM Capabilities release 中的 IFEval strict instruction-following accuracy 切片。","owner":"Stanford CRFM / IFEval","source":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLabel":"Stanford HELM · capabilities release artifact","sourceIds":["helm-capabilities"],"lastVerified":"2026-08-28","versions":[{"id":"helm-ifeval@v1.15.0","label":"HELM Capabilities v1.15.0","status":"active"}],"metrics":[{"id":"ifeval-strict-acc","label":"IFEval strict accuracy","unit":"fraction","scale":{"min":0,"max":1},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"IFEval 用可自动检查的显式约束测试指令遵循，例如格式、段落数量、关键词、大小写和结构要求。","input":"包含一个或多个可验证约束的提示","output":"同时满足内容请求和全部约束的文本响应"},"dataset":{"size":"场景与评测行按 HELM Capabilities release；当前 canonical slice 为 v1.15.0，不猜固定题量","splits":["HELM Capabilities v1.15.0","Accuracy / Efficiency / General information tables"],"availability":"HELM 发布 versioned core_scenarios.json artifact"},"protocol":{"mode":"model-only / direct via HELM","grader":"默认性能指标为 IFEval Strict Acc（fraction）；Efficiency 与 #eval/#tokens 是 telemetry","tools":"固定 HELM prompt、model endpoint、解码与 strict evaluator；通常不使用外部工具","sampling":"保留 release、table、row、column 与模型配置"},"comparison":{"recommended":"同一 HELM release、strict evaluator、prompt 与解码设置下比较 ifeval-strict-acc","avoid":"不要混用 prompt-level/instruction-level、strict/loose 指标，也不要把 runtime/token/count 当性能准确率"},"source_locator":"Stanford HELM capabilities artifact（helm-capabilities）：release=v1.15.0;table=Accuracy;row=*;column=IFEval - IFEval Strict Acc"}},{"id":"helm-math","canonicalId":"helm-math","short":"HELM MATH","name":"MATH · HELM Lite","family":"reasoning","category":"math","familyLabel":"数学","evaluationMode":"direct","comparisonKey":"helm-math:helm-math@v1.13.0:equivalent-cot","metric":"equivalent-cot","metricLabel":"Equivalent (CoT)","scale":1,"unit":"fraction","direction":"higher","version":"HELM Lite v1.13.0","versionId":"helm-math@v1.13.0","description":"HELM Lite release 中的 MATH chain-of-thought equivalence 切片。","owner":"Stanford CRFM / MATH","source":"https://crfm.stanford.edu/helm/lite/latest/config.js","sourceLabel":"Stanford HELM · lite release artifact","sourceIds":["helm-lite"],"lastVerified":"2026-08-28","versions":[{"id":"helm-math@v1.13.0","label":"HELM Lite v1.13.0","status":"active"}],"metrics":[{"id":"equivalent-cot","label":"Equivalent (CoT)","unit":"fraction","scale":{"min":0,"max":1},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"MATH 竞赛数学问题的多步推理与答案等价判定，HELM Lite 以 chain-of-thought response 的等价性衡量正确性。","input":"代数、几何、数论、组合等竞赛数学题","output":"推导文本与最终数学答案"},"dataset":{"size":"场景与评测行按 HELM Lite release；当前 canonical slice 为 v1.13.0，不假定为 MATH-500 或固定抽样规模","splits":["HELM Lite v1.13.0","MATH scenario","Accuracy / Efficiency / General information tables"],"availability":"HELM versioned core_scenarios.json artifact"},"protocol":{"mode":"model-only / direct via HELM","grader":"默认性能指标 Equivalent (CoT)（fraction）；等价判定器版本属于协议","tools":"固定 HELM prompt/shot、CoT、endpoint 和解码；若启用计算工具需另列","sampling":"保留 release/table/column；#eval、tokens 与 inference time 仅作 telemetry"},"comparison":{"recommended":"同一 HELM Lite release、MATH scenario、prompt/CoT 与 equivalence grader 内比较","avoid":"不要混入 MATH-500、不同 shot/答案判定器，也不要用 telemetry 字段替代 equivalent-cot"},"source_locator":"Stanford HELM lite artifact（helm-lite）：release=v1.13.0;table=Accuracy;row=*;column=MATH - Equivalent (CoT)"}},{"id":"helm-mmlu","canonicalId":"helm-mmlu","short":"HELM MMLU","name":"MMLU · HELM Lite","family":"knowledge","category":"knowledge","familyLabel":"知识","evaluationMode":"direct","comparisonKey":"helm-mmlu:helm-mmlu@v1.13.0:em","metric":"em","metricLabel":"Exact match","scale":1,"unit":"fraction","direction":"higher","version":"HELM Lite v1.13.0","versionId":"helm-mmlu@v1.13.0","description":"HELM Lite release 中的 MMLU exact-match accuracy 切片。","owner":"Stanford CRFM / MMLU","source":"https://crfm.stanford.edu/helm/lite/latest/config.js","sourceLabel":"Stanford HELM · lite release artifact","sourceIds":["helm-lite"],"lastVerified":"2026-08-28","versions":[{"id":"helm-mmlu@v1.13.0","label":"HELM Lite v1.13.0","status":"active"}],"metrics":[{"id":"em","label":"Exact match","unit":"fraction","scale":{"min":0,"max":1},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"MMLU 多学科选择题测试人文、社会科学、STEM 与专业领域知识和推理。","input":"学科题干与候选选项","output":"选择项答案"},"dataset":{"size":"场景、subject aggregation 与评测行按 HELM Lite release；当前 canonical slice 为 v1.13.0","splits":["HELM Lite v1.13.0","MMLU scenario","subject aggregation","Accuracy / Efficiency / General information tables"],"availability":"HELM versioned core_scenarios.json artifact"},"protocol":{"mode":"model-only / direct via HELM","grader":"默认性能指标 MMLU EM（fraction）","tools":"固定 HELM prompt、shot、subject aggregation、endpoint 与答案解析；通常不使用外部工具","sampling":"保留 release/table/column；#train/#eval、tokens、truncation 和 runtime 是 telemetry"},"comparison":{"recommended":"同一 HELM Lite release、shot、subject aggregation 与 Accuracy table 内比较 em","avoid":"不要与 MMLU-Pro 或其他 harness 的 MMLU 混排，也不要把 count/runtime 字段纳入准确率矩阵"},"source_locator":"Stanford HELM lite artifact（helm-lite）：release=v1.13.0;table=Accuracy;row=*;column=MMLU - EM"}},{"id":"helm-gsm8k","canonicalId":"helm-gsm8k","short":"HELM GSM8K","name":"GSM8K · HELM Lite","family":"reasoning","category":"math","familyLabel":"数学","evaluationMode":"direct","comparisonKey":"helm-gsm8k:helm-gsm8k@v1.13.0:em","metric":"em","metricLabel":"Exact match","scale":1,"unit":"fraction","direction":"higher","version":"HELM Lite v1.13.0","versionId":"helm-gsm8k@v1.13.0","description":"HELM Lite release 中的 GSM8K exact-match accuracy 切片。","owner":"Stanford CRFM / GSM8K","source":"https://crfm.stanford.edu/helm/lite/latest/config.js","sourceLabel":"Stanford HELM · lite release artifact","sourceIds":["helm-lite"],"lastVerified":"2026-08-28","versions":[{"id":"helm-gsm8k@v1.13.0","label":"HELM Lite v1.13.0","status":"active"}],"metrics":[{"id":"em","label":"Exact match","unit":"fraction","scale":{"min":0,"max":1},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"GSM8K 小学数学文字题测试多步算术推理与最终数值答案抽取。","input":"自然语言数学应用题","output":"推理文本与最终数值答案"},"dataset":{"size":"场景与评测行按 HELM Lite release；当前 canonical slice 为 v1.13.0，不从 #eval telemetry 推断完整数据集固定规模","splits":["HELM Lite v1.13.0","GSM8K scenario","Accuracy / Efficiency / General information tables"],"availability":"HELM versioned core_scenarios.json artifact"},"protocol":{"mode":"model-only / direct via HELM","grader":"默认性能指标 GSM8K EM（fraction），依赖最终答案抽取","tools":"固定 HELM prompt、shot/CoT、endpoint、解码和答案归一化；计算器/代码工具需另列","sampling":"保留 release/table/column；tokens、runtime 与 #eval 是 telemetry"},"comparison":{"recommended":"同一 HELM Lite release、shot/CoT、答案抽取和解码设置内比较 em","avoid":"不要与其他 GSM8K harness、不同 CoT/majority vote 混排，也不要把 telemetry 当分数"},"source_locator":"Stanford HELM lite artifact（helm-lite）：release=v1.13.0;table=Accuracy;row=*;column=GSM8K - EM"}},{"id":"swebench-lite","canonicalId":"swebench-lite","short":"SWE-bench Lite","name":"SWE-bench Lite","family":"coding","category":"coding-agent","familyLabel":"代码 Agent","evaluationMode":"system","comparisonKey":"swebench-lite:swebench-lite@official:resolved","metric":"resolved","metricLabel":"Resolved","scale":100,"unit":"%","direction":"higher","version":"official Lite leaderboard variant","versionId":"swebench-lite@official","description":"SWE-bench 官方榜单的 Lite variant；Resolved 是模型、scaffold、仓库环境和测试协议的联合结果。","owner":"SWE-bench","source":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLabel":"SWE-bench · official leaderboard JSON","sourceIds":["swebench-official"],"lastVerified":"2026-08-28","versions":[{"id":"swebench-lite@official","label":"official Lite leaderboard variant","status":"active"}],"metrics":[{"id":"resolved","label":"Resolved","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"根据真实 GitHub issue 与指定仓库快照定位问题、修改代码并生成能通过测试的补丁。","input":"issue 描述、仓库 commit 与执行环境","output":"可应用并通过评测测试的 patch"},"dataset":{"size":"Lite variant 的实例、仓库与榜单内容按 SWE-bench official release；不在目录中猜固定题量","splits":["official Lite leaderboard variant","repository/system rows"],"availability":"SWE-bench 官方 leaderboard JSON 与任务/harness 公开"},"protocol":{"mode":"system / agent harness","grader":"patch application + public/hidden tests，汇总 Resolved %","tools":"scaffold、shell、文件编辑、测试、网络、timeout、repo image 与 checked 状态必须记录","sampling":"报告 rollout 数、patch selection、模型 endpoint 与失败类型"},"comparison":{"recommended":"同一 Lite release、仓库镜像、测试策略、harness、timeout 与模型 endpoint 内比较","avoid":"不要与 Verified/Pro/Multilingual/Multimodal 混排，也不要把 Resolved 标成裸模型分数"},"source_locator":"SWE-bench official leaderboard JSON（swebench-official）：leaderboards[4]=Lite;results[*]；逐行保留 scaffold、checked、date 与 logs/trajs 字段"}},{"id":"swebench-multilingual","canonicalId":"swebench-multilingual","short":"SWE-bench Multilingual","name":"SWE-bench Multilingual","family":"coding","category":"coding-agent","familyLabel":"多语言代码 Agent","evaluationMode":"system","comparisonKey":"swebench-multilingual:swebench-multilingual@official:resolved","metric":"resolved","metricLabel":"Resolved","scale":100,"unit":"%","direction":"higher","version":"official Multilingual leaderboard variant","versionId":"swebench-multilingual@official","description":"SWE-bench 官方榜单的 Multilingual variant；语言、仓库、镜像、测试与 scaffold 必须固定。","owner":"SWE-bench","source":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLabel":"SWE-bench · official leaderboard JSON","sourceIds":["swebench-official"],"lastVerified":"2026-08-28","versions":[{"id":"swebench-multilingual@official","label":"official Multilingual leaderboard variant","status":"active"}],"metrics":[{"id":"resolved","label":"Resolved","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"在多种编程语言和仓库生态中，根据真实 issue 完成定位、修改与测试闭环。","input":"多语言仓库快照、issue 与工程环境","output":"通过该仓库评测测试的 patch"},"dataset":{"size":"语言、仓库和实例构成按 SWE-bench Multilingual official release；不登记固定题量","splits":["official Multilingual leaderboard variant","language/repository slices"],"availability":"SWE-bench 官方 leaderboard JSON；任务与环境按 release 维护"},"protocol":{"mode":"system / agent harness","grader":"patch application + repository tests，汇总 Resolved %","tools":"语言 toolchain、repo image、scaffold、测试、network、timeout 与 checked 状态属于协议","sampling":"保留语言/仓库、rollout、endpoint、scaffold 与失败分类"},"comparison":{"recommended":"同一 Multilingual release、语言/仓库构成、镜像、harness 与预算内比较","avoid":"不要与英文 Verified/Lite 总分直接平均，也不要忽略不同语言生态的测试难度"},"source_locator":"SWE-bench official leaderboard JSON（swebench-official）：leaderboards[1]=Multilingual;results[*]；逐行保留 scaffold、checked 与模型/系统字段"}},{"id":"swebench-multimodal","canonicalId":"swebench-multimodal","short":"SWE-bench Multimodal","name":"SWE-bench Multimodal","family":"coding","category":"coding-agent","familyLabel":"多模态代码 Agent","evaluationMode":"system","comparisonKey":"swebench-multimodal:swebench-multimodal@official:resolved","metric":"resolved","metricLabel":"Resolved","scale":100,"unit":"%","direction":"higher","version":"official Multimodal leaderboard variant","versionId":"swebench-multimodal@official","description":"SWE-bench 官方榜单的 Multimodal variant；视觉输入管线与 GUI/browser 工具属于系统协议。","owner":"SWE-bench","source":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLabel":"SWE-bench · official leaderboard JSON","sourceIds":["swebench-official"],"lastVerified":"2026-08-28","versions":[{"id":"swebench-multimodal@official","label":"official Multimodal leaderboard variant","status":"active"}],"metrics":[{"id":"resolved","label":"Resolved","unit":"percent","scale":{"min":0,"max":100},"direction":"higher"}],"featured":false,"displayPriority":null,"profile":{"task":{"summary":"解决包含截图、界面状态或其他视觉证据的真实 GitHub issue，并提交通过测试的代码补丁。","input":"issue、仓库快照、视觉材料与工程环境","output":"可应用、通过测试且满足视觉 issue 的 patch"},"dataset":{"size":"任务、视觉资产和仓库构成按 SWE-bench Multimodal official release；不登记固定题量","splits":["official Multimodal leaderboard variant","repository/visual issue slices"],"availability":"SWE-bench 官方 leaderboard JSON；视觉输入与任务环境按 release 维护"},"protocol":{"mode":"system / multimodal agent harness","grader":"patch application + repository tests，汇总 Resolved %","tools":"视觉输入管线、分辨率、browser/GUI、repo image、scaffold、timeout 与 checked 状态必须记录","sampling":"保留 rollout、endpoint、视觉预处理、patch selection 与失败类型"},"comparison":{"recommended":"同一 Multimodal release、视觉输入管线、仓库镜像、harness、测试和预算内比较","avoid":"不要当作纯文本模型成绩，也不要与 Lite/Verified/Multilingual 或不同 GUI 工具条件混排"},"source_locator":"SWE-bench official leaderboard JSON（swebench-official）：leaderboards[5]=Multimodal;results[*]；逐行保留 visual pipeline/scaffold、checked 与日期"}}],"models":[{"id":"openai/gpt-5.6-sol@2026-07-09","canonicalId":"openai/gpt-5.6-sol@2026-07-09","familyId":"openai/gpt-5.6","name":"GPT-5.6 Sol","provider":"OpenAI","mark":"O","release":"2026-07-09","releaseDate":"2026-07-09","access":"api","status":"active","availability":"api","aliases":["gpt-5.6-sol","gpt-5.6"],"tags":["旗舰","推理","多模态","Agent"],"summary":"OpenAI GPT-5.6 reasoning tier；API alias gpt-5.6 路由到该 release。","modalities":["text","image"],"contextWindow":1050000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"tier":"sol","preview":false},"profile":{"positioning":"GPT-5.6 旗舰 reasoning tier，面向复杂专业工作。","capabilities":["复杂推理","代码","工具调用","电脑使用","视觉输入","多语言"],"best_for":["复杂研究与长程 agent","高质量代码与调试","需要最高准确度的专业任务"],"endpoint_ids":["gpt-5.6-sol","gpt-5.6"],"availability":["api","chat","codex"],"reasoning_modes":["none","low","medium","high","xhigh","max"],"architecture":null,"context_note":"API catalog 记录 1.05M context、128K max output；具体产品 surface 可能不同。","parameter_note":"官方未披露实际权重数；下方第三方估计衡量的是 IKP-effective 知识容量，不是 total / active 参数。","parameter_estimates":[{"kind":"ikp_effective","label":"IKP-effective knowledge capacity","point_b":1930,"range_b":[559,6700],"interval":"90%","basis":"attempted","confidence":"method-dependent","source_id":"src-eigenigma-ikp-frontier-2026-07-31","as_of":"2026-07-31","note":"第三方黑盒容量估计；表示与约 1.93T 平均开放模型相当的长尾知识容量，不能表述为该模型拥有 1.93T 权重。"}],"knowledge_cutoff":"2026-02-16","max_output_tokens":128000,"license":null,"caveats":["Sol/Terra/Luna 是能力与成本 tier，不应当作相同 endpoint 的平均分。","公开成绩必须同时记录 reasoning effort、工具和 harness。"],"fact_source_ids":["src-openai-models","src-openai-gpt56","src-eigenigma-ikp-frontier-2026-07-31"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-openai-gpt56"],"scores":{"gpqa-diamond":{"value":94.6,"raw_value":"94.6%","setting":"official GPT-5.6 evaluation table; no tools","sourceId":"src-openai-gpt56","sourceUrl":"https://openai.com/index/gpt-5-6/","sourceLabel":"OpenAI · GPT-5.6 launch and evaluations","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-07-09","published_at":"2026-07-09","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":0,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"official GPT-5.6 evaluation table; no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"obs-gpt56-sol-gpqa-20260827","notes":"Provider-reported value; exact reasoning budget and evaluator details remain in the linked launch table.","legacyMigrated":false},"mmmu-pro":{"value":83.0,"raw_value":"83.0%","setting":"MMMU-Pro; no tools","sourceId":"src-openai-gpt56","sourceUrl":"https://openai.com/index/gpt-5-6/","sourceLabel":"OpenAI · GPT-5.6 launch and evaluations","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-07-09","published_at":"2026-07-09","benchmark_version":"mmmu-pro@v1","version":"Pro","metric":"accuracy","unit":"%","protocol":{"shots":0,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"MMMU-Pro; no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"obs-gpt56-sol-mmmupro-no-tools-20260827","notes":"No-tools variant is preferred for the model atlas; the with-tools variant is retained as a separate run.","legacyMigrated":false}},"scoreCount":2,"catalogOnly":false,"systemRunCount":1},{"id":"openai/gpt-5@2025-08-07","canonicalId":"openai/gpt-5@2025-08-07","familyId":"openai/gpt-5","name":"GPT-5","provider":"OpenAI","mark":"O","release":"2025-08-07","releaseDate":"2025-08-07","access":"api","status":"previous","availability":"api","aliases":["gpt-5"],"tags":["上一代","推理","工具"],"summary":"Legacy MVP baseline; GPT-5 pro values remain identified in observation notes.","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GPT-5 代际 baseline，保留用于历史纵向比较。","capabilities":["推理","代码","工具调用","视觉输入"],"best_for":["与 GPT-5.4/5.5 的代际对照","历史 benchmark 时间线"],"endpoint_ids":["gpt-5"],"availability":["api","chat"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"本目录未从当前公开资料统一核验 context，保留未注明。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["旧 release 的榜单值可能使用不同 endpoint、prompt 或工具；必须查看 observation 详情。"],"fact_source_ids":["openai-gpt5"],"last_checked":"2026-08-27"},"sourceIds":["openai-gpt5"],"scores":{"gpqa-diamond":{"value":88.4,"raw_value":"88.4%","setting":"GPT-5 pro · no tools","sourceId":"openai-gpt5","sourceUrl":"https://openai.com/index/introducing-gpt-5/","sourceLabel":"OpenAI · Introducing GPT-5","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-08-07","published_at":"2025-08-07","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"GPT-5 pro · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gpt-5-gpqa-diamond","notes":"该值是 GPT-5 pro 变体，不应与普通 GPT-5 endpoint 混为同一 release。","legacyMigrated":true},"aime-2025":{"value":94.6,"raw_value":"94.6%","setting":"no tools · high","sourceId":"openai-gpt5","sourceUrl":"https://openai.com/index/introducing-gpt-5/","sourceLabel":"OpenAI · Introducing GPT-5","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-08-07","published_at":"2025-08-07","benchmark_version":"aime-2025@v1","version":"2025","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools · high"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gpt-5-aime-2025","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"mmmu":{"value":84.2,"raw_value":"84.2%","setting":"MMMU · no tools","sourceId":"openai-gpt5","sourceUrl":"https://openai.com/index/introducing-gpt-5/","sourceLabel":"OpenAI · Introducing GPT-5","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-08-07","published_at":"2025-08-07","benchmark_version":"mmmu@v1","version":"original","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gpt-5-mmmu","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":3,"catalogOnly":false,"systemRunCount":1},{"id":"anthropic/claude-opus@4","canonicalId":"anthropic/claude-opus@4","familyId":"anthropic/claude-opus","name":"Claude Opus 4","provider":"Anthropic","mark":"A","release":"2025-05-22","releaseDate":"2025-05-22","access":"api","status":"previous","availability":"api","aliases":["claude-opus-4"],"tags":["上一代","代码","推理","Agent"],"summary":"Legacy MVP baseline; extended thinking and Claude Code scaffold are recorded per observation.","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Claude Opus 4 历史旗舰，作为 Opus 系列上一代对照。","capabilities":["推理","代码","视觉输入","Agent","extended thinking"],"best_for":["代际比较","历史 Claude Code / agent 结果"],"endpoint_ids":["claude-opus-4"],"availability":["api","hosted"],"reasoning_modes":["extended thinking"],"architecture":null,"context_note":"本目录未统一核验当前 context/配额。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["历史结果经常绑定 Claude Code scaffold；不能当作裸模型结果。"],"fact_source_ids":["anthropic-claude4"],"last_checked":"2026-08-27"},"sourceIds":["anthropic-claude4"],"scores":{"gpqa-diamond":{"value":83.3,"raw_value":"83.3%","setting":"extended thinking · 64K","sourceId":"anthropic-claude4","sourceUrl":"https://www.anthropic.com/news/claude-4","sourceLabel":"Anthropic · Introducing Claude 4","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-22","published_at":"2025-05-22","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"extended thinking · 64K"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-claude-opus-4-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"aime-2025":{"value":75.5,"raw_value":"75.5%","setting":"single attempt · extended thinking","sourceId":"anthropic-claude4","sourceUrl":"https://www.anthropic.com/news/claude-4","sourceLabel":"Anthropic · Introducing Claude 4","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-22","published_at":"2025-05-22","benchmark_version":"aime-2025@v1","version":"2025","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"single attempt · extended thinking"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-claude-opus-4-aime-2025","notes":"发布表另列出并行 test-time compute 的 90.0%，此处保留单次值。","legacyMigrated":true},"mmmu":{"value":76.5,"raw_value":"76.5%","setting":"MMMU validation · extended thinking","sourceId":"anthropic-claude4","sourceUrl":"https://www.anthropic.com/news/claude-4","sourceLabel":"Anthropic · Introducing Claude 4","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-22","published_at":"2025-05-22","benchmark_version":"mmmu@v1","version":"original","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU validation · extended thinking"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-claude-opus-4-mmmu","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":3,"catalogOnly":false,"systemRunCount":2},{"id":"qwen/qwen3.8-flash-next@2026-08-26","canonicalId":"qwen/qwen3.8-flash-next@2026-08-26","familyId":"qwen/qwen3.8","name":"Qwen3.8 Flash-Next","provider":"Qwen","mark":"Q","release":"2026-08-26","releaseDate":"2026-08-26","access":"open_weights","status":"active","availability":"open_weights","aliases":["Qwen3.8-Flash-Next","qwen3.8-flash"],"tags":["Flash","效率","开放权重","代码"],"summary":"125B main / 6B active；另有 deterministic n-gram memory，不能当作 active parameters。","modalities":["text"],"contextWindow":262144,"paramsTotal":125000000000,"paramsActive":6000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false,"long_context_max":1010000},"profile":{"positioning":"Qwen3.8 Flash-Next 高吞吐 open-weight variant，强调低成本和长上下文。","capabilities":["推理","代码","长上下文","中文/多语言"],"best_for":["高吞吐代码任务","成本敏感 agent","长文档批处理"],"endpoint_ids":["Qwen/Qwen3.8-Flash-Next","qwen3.8-flash"],"availability":["open-weights","hosted-preview"],"reasoning_modes":["thinking"],"architecture":"Hybrid Gated DeltaNet + Qwen Sparse Attention + n-gram embeddings + MoE","context_note":"262,144 native；官方 model card 描述通过指定 scaling configuration 扩展到约 1.01M。","parameter_note":"125B main / 6B active；另有 51B n-gram embeddings 与 4B MTP 组件，不能混入 6B per-token active。","knowledge_cutoff":null,"max_output_tokens":null,"license":"qwen-community-1.0","caveats":["n-gram memory 是额外组件，不应加进模型 active 参数。","Flash-Next checkpoint 与 hosted qwen3.8-flash 的 serving 版本需分开记录。"],"fact_source_ids":["src-qwen38-flash","src-qwen38-repo","src-qwen38-flash-hf"],"last_checked":"2026-08-27"},"sourceIds":["src-qwen38-flash","src-qwen38-repo","src-qwen38-flash-hf"],"scores":{"gpqa-diamond":{"value":91.7,"raw_value":"91.7%","setting":"Qwen3.8-Flash-Next official table","sourceId":"src-qwen38-flash","sourceUrl":"https://qwen.ai/blog?id=qwen3.8-flash-next","sourceLabel":"Qwen · Qwen3.8-Flash-Next","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-08-26","published_at":"2026-08-26","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"Qwen3.8-Flash-Next official table"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"obs-qwen38-flash-gpqa-20260826","notes":"Provider-reported value; protocol details follow the linked Qwen release page.","legacyMigrated":false},"livecodebench":{"value":91.9,"raw_value":"91.9%","setting":"LiveCodeBench v6; Qwen official table","sourceId":"src-qwen38-flash","sourceUrl":"https://qwen.ai/blog?id=qwen3.8-flash-next","sourceLabel":"Qwen · Qwen3.8-Flash-Next","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-08-26","published_at":"2026-08-26","benchmark_version":"livecodebench@v6","version":"V6","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"LiveCodeBench v6; Qwen official table"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"obs-qwen38-flash-lcb-20260826","notes":"Provider-reported rolling benchmark value; window and evaluator should be checked before exact comparison.","legacyMigrated":false}},"scoreCount":2,"catalogOnly":false,"systemRunCount":1},{"id":"qwen/qwen3-max@thinking","canonicalId":"qwen/qwen3-max@thinking","familyId":"qwen/qwen3-max","name":"Qwen3-Max-Thinking","provider":"Qwen","mark":"Q","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["qwen3-max-thinking","qwen3-max-2026-01-23"],"tags":["上一代","推理","代码","多语言"],"summary":"Legacy seed model；保留用于纵向比较。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3-Max-Thinking 历史 reasoning 服务 release。","capabilities":["深度推理","代码","工具调用","中文/多语言"],"best_for":["Qwen reasoning 时间线","复杂数学/代码历史对照"],"endpoint_ids":["qwen3-max-2026-01-23","qwen3-max-thinking"],"availability":["api","hosted"],"reasoning_modes":["thinking"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["Thinking 是运行模式/endpoint 标识，不应和 Qwen3-Max-Instruct 混合。"],"fact_source_ids":["qwen3-max"],"last_checked":"2026-08-27"},"sourceIds":["qwen3-max"],"scores":{"gpqa-diamond":{"value":87.4,"raw_value":"87.4%","setting":"官方横向表 · no custom scaffold","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-01-15","published_at":"2026-01-15","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"官方横向表 · no custom scaffold"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-qwen3-max-thinking-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"livecodebench":{"value":85.9,"raw_value":"85.9%","setting":"LiveCodeBench v6","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-01-15","published_at":"2026-01-15","benchmark_version":"livecodebench@v5","version":"V5","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"LiveCodeBench v6"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-qwen3-max-thinking-livecodebench","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"mmlu-pro":{"value":85.7,"raw_value":"85.7%","setting":"官方横向表","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-01-15","published_at":"2026-01-15","benchmark_version":"mmlu-pro@v1","version":"v1","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"官方横向表"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-qwen3-max-thinking-mmlu-pro","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"hle":{"value":30.2,"raw_value":"30.2%","setting":"text subset · no tools","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-01-15","published_at":"2026-01-15","benchmark_version":"hle@v1","version":"full","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"text subset · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-qwen3-max-thinking-hle","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":4,"catalogOnly":false,"systemRunCount":1},{"id":"deepseek/r1@0528","canonicalId":"deepseek/r1@0528","familyId":"deepseek/r1","name":"DeepSeek-R1-0528","provider":"DeepSeek","mark":"D","release":"2025-05-28","releaseDate":"2025-05-28","access":"open_weights","status":"previous","availability":"open_weights","aliases":["deepseek-r1-0528"],"tags":["历史","推理","开放权重"],"summary":"","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"DeepSeek-R1-0528 历史 reasoning release，保留作开放权重推理基线。","capabilities":["深度推理","数学","代码"],"best_for":["数学/推理时间线","开放权重 reasoning 对照"],"endpoint_ids":["DeepSeek-R1-0528"],"availability":["open-weights"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["模型卡、量化和推理框架不同会改变速度/显存；benchmark 仍需保留 protocol。"],"fact_source_ids":["deepseek-r1-0528"],"last_checked":"2026-08-27"},"sourceIds":["deepseek-r1-0528"],"scores":{"gpqa-diamond":{"value":68.4,"raw_value":"68.4%","setting":"R1-0528 · reported","sourceId":"deepseek-r1-0528","sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API · R1-0528 update","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-28","published_at":"2025-05-28","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"R1-0528 · reported"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-deepseek-r1-0528-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"aime-2025":{"value":87.5,"raw_value":"87.5%","setting":"R1-0528 · reported","sourceId":"deepseek-r1-0528","sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API · R1-0528 update","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-28","published_at":"2025-05-28","benchmark_version":"aime-2025@v1","version":"2025","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"R1-0528 · reported"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-deepseek-r1-0528-aime-2025","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"mmlu-pro":{"value":81.2,"raw_value":"81.2%","setting":"R1-0528 · reported","sourceId":"deepseek-r1-0528","sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API · R1-0528 update","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-28","published_at":"2025-05-28","benchmark_version":"mmlu-pro@v1","version":"v1","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"R1-0528 · reported"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-deepseek-r1-0528-mmlu-pro","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":3,"catalogOnly":false,"systemRunCount":0},{"id":"google/gemini-3.1-pro@2026-02-19","canonicalId":"google/gemini-3.1-pro@2026-02-19","familyId":"google/gemini-3.1","name":"Gemini 3.1 Pro","provider":"Google","mark":"G","release":"2026-02-19","releaseDate":"2026-02-19","access":"api","status":"preview","availability":"api","aliases":["gemini-3.1-pro","gemini-3-1-pro-preview"],"tags":["多模态","推理","Agent"],"summary":"Gemini 3.1 Pro preview endpoint；1,048,576-token input、65,536-token max output，多模态原生能力。","modalities":["text","image","video","audio"],"contextWindow":1048576,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":true},"profile":{"positioning":"Gemini 3.1 Pro 多模态 reasoning baseline，覆盖长上下文和 agent 场景。","capabilities":["推理","文本","图像","视频","音频","工具调用"],"best_for":["多模态知识工作","Google 生态工具任务","跨模态 reasoning"],"endpoint_ids":["gemini-3.1-pro","gemini-3-1-pro-preview"],"availability":["api","hosted"],"reasoning_modes":["thinking"],"architecture":"Native multimodal model based on Gemini 3 Pro（官方高层描述）","context_note":"官方 preview endpoint 规格记录 1,048,576 input tokens。","parameter_note":"参数量未公开。","max_output_tokens":65536,"license":null,"caveats":["Preview alias 与稳定 endpoint 可能路由不同；保留 source snapshot 和日期。"],"fact_source_ids":["google-gemini31","src-google-gemini31-spec"],"last_checked":"2026-08-27"},"sourceIds":["google-gemini31","src-google-gemini31-spec"],"scores":{"gpqa-diamond":{"value":94.3,"raw_value":"94.3%","setting":"no tools · thinking high","sourceId":"google-gemini31","sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","sourceLabel":"Google DeepMind · Gemini 3.1 Pro model card","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-02-19","published_at":"2026-02-19","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools · thinking high"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-3-1-pro-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"mmmu":{"value":80.5,"raw_value":"80.5%","setting":"MMMU-Pro · no tools","sourceId":"google-gemini31","sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","sourceLabel":"Google DeepMind · Gemini 3.1 Pro model card","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-02-19","published_at":"2026-02-19","benchmark_version":"mmmu@v1","version":"original","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU-Pro · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-3-1-pro-mmmu","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"hle":{"value":44.4,"raw_value":"44.4%","setting":"full set · no tools","sourceId":"google-gemini31","sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","sourceLabel":"Google DeepMind · Gemini 3.1 Pro model card","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-02-19","published_at":"2026-02-19","benchmark_version":"hle@v1","version":"full","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"full set · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-3-1-pro-hle","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":3,"catalogOnly":false,"systemRunCount":2},{"id":"google/gemini-2.5-pro@2025-06-05","canonicalId":"google/gemini-2.5-pro@2025-06-05","familyId":"google/gemini-2.5","name":"Gemini 2.5 Pro","provider":"Google","mark":"G","release":"2025-06-05","releaseDate":"2025-06-05","access":"api","status":"previous","availability":"api","aliases":["gemini-2.5-pro"],"tags":["上一代","多模态","长上下文"],"summary":"","modalities":["text","image","video","audio"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Gemini 2.5 Pro 上一代多模态 reasoning baseline。","capabilities":["推理","文本","图像","视频","音频","长上下文"],"best_for":["Gemini 3.1 代际比较","多模态历史 benchmark"],"endpoint_ids":["gemini-2.5-pro"],"availability":["api","hosted"],"reasoning_modes":["thinking"],"architecture":null,"context_note":"当前目录未统一核验 context；不同 Google API surface 可能有不同上限。","parameter_note":"参数量未公开。","license":null,"caveats":["Gemini benchmark 常区分 thinking budget、tools 和 modality；不可只按显示名合并。"],"fact_source_ids":["google-gemini25"],"last_checked":"2026-08-27"},"sourceIds":["google-gemini25"],"scores":{"gpqa-diamond":{"value":86.4,"raw_value":"86.4%","setting":"single attempt · no tools","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-06-05","published_at":"2025-06-05","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"single attempt · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-2-5-pro-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"aime-2025":{"value":88.0,"raw_value":"88.0%","setting":"single attempt · no tools","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-06-05","published_at":"2025-06-05","benchmark_version":"aime-2025@v1","version":"2025","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"single attempt · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-2-5-pro-aime-2025","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"livecodebench":{"value":70.4,"raw_value":"70.4%","setting":"V5 · rolling window","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-03-25","published_at":"2025-03-25","benchmark_version":"livecodebench@v5","version":"V5","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"V5 · rolling window"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-2-5-pro-livecodebench","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"mmmu":{"value":81.7,"raw_value":"81.7%","setting":"MMMU · single attempt","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-06-05","published_at":"2025-06-05","benchmark_version":"mmmu@v1","version":"original","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU · single attempt"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-2-5-pro-mmmu","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"hle":{"value":18.8,"raw_value":"18.8%","setting":"no tools","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-03-25","published_at":"2025-03-25","benchmark_version":"hle@v1","version":"full","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-2-5-pro-hle","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":5,"catalogOnly":false,"systemRunCount":1},{"id":"xai/grok-4-fast@2025-09-19","canonicalId":"xai/grok-4-fast@2025-09-19","familyId":"xai/grok-4","name":"Grok 4 Fast","provider":"xAI","mark":"X","release":"2025-09-19","releaseDate":"2025-09-19","access":"api","status":"previous","availability":"api","aliases":["grok-4-fast"],"tags":["上一代","实时","推理","效率"],"summary":"","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"speed_variant":true},"profile":{"positioning":"Grok 4 Fast 高吞吐、实时导向的上一代 serving tier。","capabilities":["推理","代码","实时信息","视觉输入"],"best_for":["实时问答","高吞吐 API","Grok 代际速度比较"],"endpoint_ids":["grok-4-fast"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","license":null,"caveats":["实时检索和工具开关会改变结果；需要区分裸模型与带 search 的 system。"],"fact_source_ids":["xai-grok4fast"],"last_checked":"2026-08-27"},"sourceIds":["xai-grok4fast"],"scores":{"gpqa-diamond":{"value":85.7,"raw_value":"85.7%","setting":"no tools","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-09-19","published_at":"2025-09-19","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-grok-4-fast-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"aime-2025":{"value":92.0,"raw_value":"92.0%","setting":"no tools","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-09-19","published_at":"2025-09-19","benchmark_version":"aime-2025@v1","version":"2025","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-grok-4-fast-aime-2025","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"livecodebench":{"value":80.0,"raw_value":"80.0%","setting":"Jan–May rolling window","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-09-19","published_at":"2025-09-19","benchmark_version":"livecodebench@v5","version":"V5","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"Jan–May rolling window"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-grok-4-fast-livecodebench","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"hle":{"value":20.0,"raw_value":"20.0%","setting":"no tools","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-09-19","published_at":"2025-09-19","benchmark_version":"hle@v1","version":"full","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-grok-4-fast-hle","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":4,"catalogOnly":false,"systemRunCount":0}],"sources":[{"id":"openai-gpt5","label":"OpenAI · Introducing GPT-5","kind":"primary","publisher":"OpenAI","url":"https://openai.com/index/introducing-gpt-5/","default_evidence_level":"B","staleness_after_days":180},{"id":"anthropic-claude4","label":"Anthropic · Introducing Claude 4","kind":"primary","publisher":"Anthropic","url":"https://www.anthropic.com/news/claude-4","default_evidence_level":"B","staleness_after_days":180},{"id":"google-gemini25","label":"Google · Gemini 2.5 model card/blog","kind":"primary","publisher":"Google DeepMind","url":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","default_evidence_level":"B","staleness_after_days":180},{"id":"google-gemini31","label":"Google DeepMind · Gemini 3.1 Pro model card","kind":"primary","publisher":"Google DeepMind","url":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","default_evidence_level":"B","staleness_after_days":120},{"id":"xai-grok4fast","label":"xAI · Grok 4 Fast","kind":"primary","publisher":"xAI","url":"https://x.ai/news/grok-4-fast","default_evidence_level":"B","staleness_after_days":180},{"id":"deepseek-r1-0528","label":"DeepSeek API · R1-0528 update","kind":"primary","publisher":"DeepSeek","url":"https://api-docs.deepseek.com/updates/","default_evidence_level":"B","staleness_after_days":180},{"id":"qwen3-max","label":"Qwen · Qwen3-Max-Thinking","kind":"primary","publisher":"Qwen","url":"https://qwen.ai/blog?id=qwen3-max-thinking","default_evidence_level":"B","staleness_after_days":180},{"id":"src-openai-models","label":"OpenAI API model catalog","kind":"documentation","publisher":"OpenAI","url":"https://developers.openai.com/api/docs/models","default_evidence_level":"A","staleness_after_days":30},{"id":"src-openai-gpt56","label":"OpenAI · GPT-5.6 launch and evaluations","kind":"primary","publisher":"OpenAI","url":"https://openai.com/index/gpt-5-6/","default_evidence_level":"B","staleness_after_days":120},{"id":"src-openai-gpt55","label":"OpenAI · Introducing GPT-5.5","kind":"primary","publisher":"OpenAI","url":"https://openai.com/index/introducing-gpt-5-5/","default_evidence_level":"B","staleness_after_days":180},{"id":"src-anthropic-opus5","label":"Anthropic · Claude Opus 5","kind":"primary","publisher":"Anthropic","url":"https://www.anthropic.com/news/claude-opus-5","default_evidence_level":"B","staleness_after_days":120},{"id":"src-anthropic-sonnet5","label":"Anthropic · Claude Sonnet 5 research","kind":"primary","publisher":"Anthropic","url":"https://www.anthropic.com/research/claude-sonnet-5","default_evidence_level":"B","staleness_after_days":120},{"id":"src-anthropic-fable5","label":"Anthropic · Claude Fable 5 and Mythos 5","kind":"primary","publisher":"Anthropic","url":"https://www.anthropic.com/news/claude-fable-5-mythos-5","default_evidence_level":"B","staleness_after_days":120},{"id":"src-anthropic-system-cards","label":"Anthropic system cards index","kind":"documentation","publisher":"Anthropic","url":"https://www.anthropic.com/system-cards","default_evidence_level":"A","staleness_after_days":90},{"id":"src-qwen38-flash","label":"Qwen · Qwen3.8-Flash-Next","kind":"primary","publisher":"Qwen","url":"https://qwen.ai/blog?id=qwen3.8-flash-next","default_evidence_level":"B","staleness_after_days":90},{"id":"src-qwen38-repo","label":"Qwen3.8 official repository","kind":"primary","publisher":"Qwen","url":"https://github.com/QwenLM/Qwen3.8","default_evidence_level":"A","staleness_after_days":90},{"id":"src-qwen38-27b","label":"Qwen · Qwen3.8-27B","kind":"primary","publisher":"Qwen","url":"https://qwen.ai/blog?id=qwen3.8-27b","default_evidence_level":"B","staleness_after_days":90},{"id":"src-qwen36","label":"Qwen · Qwen3.6 release notes","kind":"primary","publisher":"Qwen","url":"https://qwen.ai/blog?id=qwen3.6-35b-a3b","default_evidence_level":"B","staleness_after_days":180},{"id":"src-deepseek-v4","label":"DeepSeek V4 preview release","kind":"primary","publisher":"DeepSeek","url":"https://api-docs.deepseek.com/news/news260424/","default_evidence_level":"B","staleness_after_days":90},{"id":"src-deepseek-updates","label":"DeepSeek API change log","kind":"primary","publisher":"DeepSeek","url":"https://api-docs.deepseek.com/updates/","default_evidence_level":"B","staleness_after_days":60},{"id":"src-kimi-k3","label":"Kimi · K3 technical blog","kind":"primary","publisher":"Moonshot AI","url":"https://www.kimi.com/en/blog/kimi-k3","default_evidence_level":"B","staleness_after_days":90},{"id":"src-kimi-code","label":"Kimi Code model configuration","kind":"documentation","publisher":"Moonshot AI","url":"https://www.kimi.com/code/docs/en/kimi-code/models.html","default_evidence_level":"A","staleness_after_days":30},{"id":"src-glm53","label":"Z.ai · GLM-5.3","kind":"primary","publisher":"Z.ai","url":"https://z.ai/blog/glm-5.3","default_evidence_level":"B","staleness_after_days":90},{"id":"src-glm53-flash","label":"Z.ai · GLM-5.3 Flash","kind":"primary","publisher":"Z.ai","url":"https://z.ai/blog/glm-5.3-flash","default_evidence_level":"B","staleness_after_days":90},{"id":"src-glm52","label":"Z.ai · GLM-5.2","kind":"primary","publisher":"Z.ai","url":"https://z.ai/blog/glm-5.2","default_evidence_level":"B","staleness_after_days":180},{"id":"src-minimax-release","label":"MiniMax model release notes","kind":"documentation","publisher":"MiniMax","url":"https://platform.minimaxi.com/docs/release-notes/models","default_evidence_level":"A","staleness_after_days":45},{"id":"src-minimax-m3","label":"MiniMax · M3","kind":"primary","publisher":"MiniMax","url":"https://www.minimax.io/models/text/m3","default_evidence_level":"B","staleness_after_days":120},{"id":"src-mimo-models","label":"MiMo current model catalog","kind":"documentation","publisher":"Xiaomi MiMo","url":"https://mimo.mi.com/docs/en-US/quick-start/model","default_evidence_level":"A","staleness_after_days":45},{"id":"src-mimo-updates","label":"MiMo model updates","kind":"documentation","publisher":"Xiaomi MiMo","url":"https://mimo.mi.com/docs/en-US/updates/model","default_evidence_level":"A","staleness_after_days":60},{"id":"gpqa-paper","label":"GPQA paper","kind":"benchmark","publisher":"Stanford / Scale","url":"https://arxiv.org/abs/2311.12022","default_evidence_level":"A","staleness_after_days":730},{"id":"swebench","label":"SWE-bench official leaderboard","kind":"benchmark","publisher":"SWE-bench","url":"https://www.swebench.com/","default_evidence_level":"A","staleness_after_days":30},{"id":"swebench-official","label":"SWE-bench · official leaderboard JSON","kind":"benchmark","publisher":"SWE-bench","url":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","web_url":"https://www.swebench.com/","api_url":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","default_evidence_level":"A","staleness_after_days":30,"adapter_id":"swebench-official","enabled":true,"notes":"Official JSON leaderboard; preserves benchmark variant, agent/scaffold, checked flag, cost and instance calls. Rows are source-reported and not independently reproduced."},{"id":"livecodebench","label":"LiveCodeBench official leaderboard","kind":"benchmark","publisher":"LiveCodeBench","url":"https://livecodebench.github.io/leaderboard.html","default_evidence_level":"A","staleness_after_days":45,"legacy_url":"https://livecodebench.github.io/leaderboard_v5.html","notes":"动态页面默认展示当前时间窗口；历史 V5 页面保留为 legacy_url。记录 start/end dates、release version、语言与污染标记。"},{"id":"mmlu-pro","label":"MMLU-Pro repository","kind":"benchmark","publisher":"TIGER-AI-Lab","url":"https://github.com/TIGER-AI-Lab/MMLU-Pro","default_evidence_level":"A","staleness_after_days":730},{"id":"mmmu","label":"MMMU benchmark","kind":"benchmark","publisher":"MMMU","url":"https://mmmu-benchmark.github.io/","default_evidence_level":"A","staleness_after_days":730},{"id":"terminal-bench","label":"Terminal-Bench","kind":"benchmark","publisher":"Terminal-Bench","url":"https://www.tbench.ai/","default_evidence_level":"A","staleness_after_days":30},{"id":"hle","label":"Scale Labs HLE leaderboard","kind":"benchmark","publisher":"Scale AI","url":"https://labs.scale.com/leaderboard/humanitys_last_exam","default_evidence_level":"A","staleness_after_days":60},{"id":"src-frontiermath","label":"FrontierMath benchmark","kind":"benchmark","publisher":"Epoch AI","url":"https://epoch.ai/benchmarks/frontiermath","default_evidence_level":"A","staleness_after_days":180},{"id":"src-frontiermath-v2","label":"Epoch AI · FrontierMath Tier 4 v2","kind":"benchmark","publisher":"Epoch AI","url":"https://epoch.ai/benchmarks/frontiermath-tier-4-v2","default_evidence_level":"A","staleness_after_days":90,"notes":"Official v2 benchmark page; records the 2026-06-12 release, corrected task set, Tier 4 protocol, Python tool and 1M-token hard limit."},{"id":"src-hmmt","label":"HMMT benchmark","kind":"benchmark","publisher":"HMMT","url":"https://www.hmmt.org/","default_evidence_level":"A","staleness_after_days":365},{"id":"src-maa-aime","label":"MAA · American Invitational Mathematics Examination","kind":"benchmark","publisher":"Mathematical Association of America","url":"https://maa.org/maa-invitational-competitions/","default_evidence_level":"A","staleness_after_days":180,"notes":"Official AIME competition source. AIME I/II, publication window, tool access and pass@1 versus best-of-n remain protocol fields; HMMT is a different competition."},{"id":"src-simpleqa","label":"SimpleQA benchmark","kind":"benchmark","publisher":"OpenAI","url":"https://github.com/openai/simple-evals","default_evidence_level":"A","staleness_after_days":365},{"id":"src-swebench-pro","label":"SWE-bench Pro paper","kind":"benchmark","publisher":"SWE-bench","url":"https://arxiv.org/abs/2509.16941","default_evidence_level":"A","staleness_after_days":365},{"id":"src-deepswe","label":"DeepSWE leaderboard","kind":"benchmark","publisher":"DataCurve","url":"https://deepswe.datacurve.ai/","default_evidence_level":"C","staleness_after_days":45},{"id":"src-nl2repo","label":"NL2Repo benchmark","kind":"benchmark","publisher":"M-A-P / NL2Repo-Bench","url":"https://github.com/multimodal-art-projection/NL2RepoBench","default_evidence_level":"A","staleness_after_days":365},{"id":"src-bfcl","label":"Berkeley Function Calling Leaderboard","kind":"official_artifact","publisher":"Berkeley Gorilla Lab","url":"https://gorilla.cs.berkeley.edu/leaderboard","api_url":"https://gorilla.cs.berkeley.edu/data_overall.csv","download_url":"https://gorilla.cs.berkeley.edu/data_overall.csv","default_evidence_level":"A","staleness_after_days":60,"adapter_id":"src-bfcl","license_note":"Official BFCL V4 aggregate CSV; retain model link, calling mode, evaluator commit and category-level fields."},{"id":"src-taubench","label":"τ-bench","kind":"benchmark","publisher":"τ-bench","url":"https://taubench.com/","default_evidence_level":"A","staleness_after_days":365},{"id":"src-toolathlon","label":"Toolathlon benchmark","kind":"benchmark","publisher":"Toolathlon","url":"https://toolathlon.ai/","default_evidence_level":"A","staleness_after_days":120},{"id":"src-gdpval","label":"GDPval methodology","kind":"benchmark","publisher":"Artificial Analysis","url":"https://artificialanalysis.ai/methodology/intelligence-benchmarking","default_evidence_level":"C","staleness_after_days":90},{"id":"src-osworld","label":"OSWorld-V2 benchmark and release manifest","kind":"benchmark","publisher":"OSWorld","url":"https://github.com/xlang-ai/OSWorld-V2","default_evidence_level":"A","staleness_after_days":90,"notes":"Official OSWorld-V2 repository; pin code, tasks, assets, websites, provider images and the osworld-v2-2026.06.24 release manifest together."},{"id":"src-helm","label":"Stanford HELM","kind":"benchmark","publisher":"Stanford CRFM","url":"https://crfm.stanford.edu/helm/index.html","default_evidence_level":"A","staleness_after_days":90},{"id":"src-livebench","label":"LiveBench","kind":"benchmark","publisher":"LiveBench","url":"https://livebench.ai/","default_evidence_level":"A","staleness_after_days":120},{"id":"src-ceval","label":"C-Eval repository","kind":"benchmark","publisher":"HKUST NLP","url":"https://github.com/hkust-nlp/ceval","default_evidence_level":"A","staleness_after_days":730},{"id":"src-cmmlu","label":"CMMLU repository","kind":"benchmark","publisher":"CMMLU","url":"https://github.com/haonan-li/CMMLU","default_evidence_level":"A","staleness_after_days":730},{"id":"src-superclue","label":"SuperCLUE arena","kind":"benchmark","publisher":"SuperCLUE","url":"https://www.superclueai.com/arena","default_evidence_level":"C","staleness_after_days":60},{"id":"src-mmmlu","label":"MMMLU dataset","kind":"benchmark","publisher":"OpenAI","url":"https://huggingface.co/datasets/openai/MMMLU","default_evidence_level":"A","staleness_after_days":365},{"id":"src-cybergym","label":"CyberGym benchmark","kind":"benchmark","publisher":"CyberGym","url":"https://cybergym.ai/","default_evidence_level":"C","staleness_after_days":120},{"id":"hf-swebench-verified","label":"Hugging Face · SWE-bench Verified leaderboard API","kind":"official_api","publisher":"Hugging Face / SWE-bench","url":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","default_evidence_level":"A","staleness_after_days":14,"adapter_id":"hf-swebench-verified","license_note":"Store candidate metadata and hashes; verify dataset/model-card redistribution terms before saving payload."},{"id":"hf-swebench-pro","label":"Hugging Face · SWE-bench Pro leaderboard API","kind":"official_api","publisher":"Hugging Face / Scale","url":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","default_evidence_level":"A","staleness_after_days":30,"adapter_id":"hf-swebench-pro"},{"id":"hf-hle","label":"Hugging Face · Humanity's Last Exam leaderboard API","kind":"official_api","publisher":"Hugging Face / HLE","url":"https://huggingface.co/api/datasets/cais/hle/leaderboard","default_evidence_level":"A","staleness_after_days":30,"adapter_id":"hf-hle"},{"id":"hf-mmlu-pro","label":"Hugging Face · MMLU-Pro leaderboard API","kind":"official_api","publisher":"Hugging Face / TIGER-Lab","url":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","default_evidence_level":"A","staleness_after_days":30,"adapter_id":"hf-mmlu-pro"},{"id":"hf-gpqa","label":"Hugging Face · GPQA leaderboard API","kind":"official_api","publisher":"Hugging Face / GPQA","url":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","default_evidence_level":"A","staleness_after_days":90,"adapter_id":"hf-gpqa"},{"id":"hf-terminal-bench","label":"Hugging Face · Terminal-Bench leaderboard API","kind":"official_api","publisher":"Hugging Face / Harbor","url":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","default_evidence_level":"A","staleness_after_days":30,"adapter_id":"hf-terminal-bench"},{"id":"hf-aime-2026","label":"Hugging Face · AIME leaderboard API","kind":"official_api","publisher":"Hugging Face / MathArena","url":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","default_evidence_level":"A","staleness_after_days":30,"adapter_id":"hf-aime-2026"},{"id":"hf-hmmt-2026","label":"Hugging Face · HMMT leaderboard API","kind":"official_api","publisher":"Hugging Face / MathArena","url":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","default_evidence_level":"A","staleness_after_days":30,"adapter_id":"hf-hmmt-2026"},{"id":"livebench-official","label":"LiveBench · official dated release table","kind":"official_repository","publisher":"LiveBench","url":"https://github.com/LiveBench/new-livebench","api_url":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","default_evidence_level":"A","staleness_after_days":14,"adapter_id":"livebench-official"},{"id":"helm-capabilities","label":"Stanford HELM · capabilities release artifact","kind":"official_artifact","publisher":"Stanford CRFM","url":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","default_evidence_level":"A","staleness_after_days":90,"adapter_id":"helm-capabilities"},{"id":"helm-lite","label":"Stanford HELM · lite release artifact","kind":"official_artifact","publisher":"Stanford CRFM","url":"https://crfm.stanford.edu/helm/lite/latest/config.js","default_evidence_level":"A","staleness_after_days":90,"adapter_id":"helm-lite"},{"id":"lmsys-arena","label":"LMArena / Chatbot Arena · metadata only","kind":"unstable_web_app","publisher":"LMArena / LMSYS","url":"https://arena.ai/leaderboard","default_evidence_level":"C","staleness_after_days":14,"adapter_id":"lmsys-arena","enabled":false,"license_note":"No undocumented endpoint scraping; add a dated, auditable export before ingesting Elo."},{"id":"lmarena-hf-dataset","label":"Arena · official Hugging Face leaderboard dataset","kind":"official_dataset_api","publisher":"Arena / LMArena","url":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","api_url":"https://datasets-server.huggingface.co/rows","default_evidence_level":"A","staleness_after_days":14,"adapter_id":"lmarena-hf-dataset","enabled":true,"license":"cc-by-4.0","license_note":"Arena dataset card declares CC-BY-4.0; preserve attribution and retain source locator."},{"id":"agents-last-exam","label":"Agents' Last Exam · ALE-V1 leaderboard","kind":"official_api","publisher":"UC Berkeley RDI","url":"https://agents-last-exam.org/leaderboard","api_url":"https://agents-last-exam.org/api/demo/leaderboard","default_evidence_level":"A","staleness_after_days":14,"adapter_id":"agents-last-exam","enabled":true,"license":"cc-by-4.0","license_note":"ALE site states the dataset is CC BY 4.0; retain endpoint, row locator, split and harness attribution."},{"id":"src-ale-bench","label":"SakanaAI · ALE-Bench leaderboard artifact (catalog only)","kind":"official_artifact","publisher":"SakanaAI","url":"https://sakanaai.github.io/ALE-Bench-Leaderboard/","api_url":"https://sakanaai.github.io/ALE-Bench-Leaderboard/data/results_summary.json","download_url":"https://sakanaai.github.io/ALE-Bench-Leaderboard/download_results_json.sh","default_evidence_level":"A","staleness_after_days":120,"enabled":false,"license":"cc-by-nd-4.0","license_note":"Catalog-only for now; leaderboard data is CC BY-ND 4.0 and must be checked before redistribution or automated ingestion.","notes":"Distinct from Agents' Last Exam (ALE-V1): this ALE-Bench measures long-horizon algorithm engineering on AtCoder Heuristic Contests."},{"id":"src-browsecomp","label":"OpenAI · BrowseComp benchmark","kind":"benchmark","publisher":"OpenAI","url":"https://openai.com/index/browsecomp/","default_evidence_level":"A","staleness_after_days":180,"enabled":false,"notes":"Official methodology/task release; no documented public leaderboard API. Keep provider-reported scores as candidate evidence with agent/tool setting and retrieval date."},{"id":"src-aider-polyglot","label":"Aider · Polyglot coding leaderboard","kind":"official_repository","publisher":"Aider","url":"https://aider.chat/docs/leaderboards/","api_url":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","download_url":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","default_evidence_level":"A","staleness_after_days":30,"enabled":true,"adapter_id":"src-aider-polyglot","notes":"Official repository YAML is machine-readable; page remains the human-facing context. Preserve the raw commit URL, edit format and date; rows are reported candidates, not reproduced runs."},{"id":"src-androidworld","label":"Google Research · AndroidWorld","kind":"official_repository","publisher":"Google Research","url":"https://github.com/google-research/android_world","default_evidence_level":"A","staleness_after_days":180,"enabled":false,"notes":"Repository documents tasks and evaluation; no stable leaderboard API. Results are system-level and must retain Android emulator/image, harness and task version."},{"id":"src-mle-bench","label":"OpenAI · MLE-bench","kind":"official_repository","publisher":"OpenAI","url":"https://github.com/openai/mle-bench/","api_url":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","download_url":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","default_evidence_level":"A","staleness_after_days":180,"enabled":true,"adapter_id":"src-mle-bench","notes":"Official repository README contains a versioned Markdown leaderboard table; rows are system candidates. Preserve grading report, environment, complexity split and date. New submissions are currently paused by the project."},{"id":"src-epoch-benchmark-hub","label":"Epoch AI · Benchmarking Hub downloadable snapshot","kind":"official_artifact","publisher":"Epoch AI","url":"https://epoch.ai/benchmarks/use-this-data","api_url":"https://epoch.ai/data/benchmark_data.zip","download_url":"https://epoch.ai/data/benchmark_data.zip","default_evidence_level":"A","staleness_after_days":14,"adapter_id":"src-epoch-benchmark-hub","enabled":true,"license":"CC-BY-4.0","license_note":"Epoch AI permits reuse/distribution with attribution; external tables retain original licences. Keep source file, row locator, and original source link."},{"id":"src-mathvista","label":"MathVista · official project and dataset","kind":"benchmark","publisher":"MathVista","url":"https://mathvista.github.io/","default_evidence_level":"A","staleness_after_days":365,"notes":"Official project page documents the 6,141-example dataset, testmini/test split, task families and evaluation protocol."},{"id":"src-video-mme","label":"Video-MME · official project","kind":"benchmark","publisher":"Video-MME","url":"https://video-mme.github.io/","default_evidence_level":"A","staleness_after_days":365,"notes":"Official project page and paper; preserve video release, duration bucket, frame sampling, audio and subtitle settings."},{"id":"src-ruler","label":"RULER · official code and paper","kind":"benchmark","publisher":"NVIDIA / RULER authors","url":"https://github.com/NVIDIA/RULER","default_evidence_level":"A","staleness_after_days":365,"notes":"Official repository generates configurable synthetic tasks; version, seed, task configuration and context buckets are part of the protocol."},{"id":"src-longbench","label":"LongBench v2 · official project","kind":"benchmark","publisher":"LongBench","url":"https://longbench2.github.io/","default_evidence_level":"A","staleness_after_days":365,"notes":"Official LongBench v2 project; retain the v2 question set, six task categories, language and context-length settings."},{"id":"src-cybergym-paper","label":"CyberGym · paper and evaluator","kind":"benchmark","publisher":"UC Berkeley RDI","url":"https://arxiv.org/abs/2506.02548","default_evidence_level":"A","staleness_after_days":365,"notes":"Paper defines the vulnerability-reproduction task and reports the 1,507-instance release; environment/scorer versions remain required."},{"id":"src-qwen35","label":"Qwen · Qwen3.5 model family","kind":"primary","publisher":"Qwen","url":"https://qwen.ai/blog?id=qwen3.5","default_evidence_level":"A","staleness_after_days":180,"notes":"Official family announcement and model cards; preserve the exact dense/MoE checkpoint because Qwen3.5 sizes are separate releases."},{"id":"src-qwen35-max-preview","label":"Qwen · Qwen3.5-Max-Preview Arena announcement","kind":"primary","publisher":"Qwen","url":"https://qwen.ai/blog?id=qwen3.5-max-preview","default_evidence_level":"B","staleness_after_days":90,"notes":"Official preview announcement dated 2026-03-18; it identifies an Arena preview, not a final or open-weight Max release."},{"id":"src-qwen3-coder","label":"Qwen · Qwen3-Coder 480B-A35B","kind":"primary","publisher":"Qwen","url":"https://qwenlm.github.io/blog/qwen3-coder/","default_evidence_level":"A","staleness_after_days":365,"notes":"Official announcement for Qwen3-Coder-480B-A35B-Instruct; exact checkpoint, context extension and agent harness must remain visible."},{"id":"src-qwen3-coder-next","label":"Qwen · Qwen3-Coder-Next model card","kind":"primary","publisher":"Qwen","url":"https://huggingface.co/Qwen/Qwen3-Coder-Next","default_evidence_level":"A","staleness_after_days":180,"notes":"Official Qwen organization model card; Next is a distinct checkpoint and is not collapsed into the 480B-A35B release."},{"id":"src-google-gemini-models","label":"Google · Gemini API model catalog","kind":"documentation","publisher":"Google DeepMind","url":"https://ai.google.dev/gemini-api/docs/models","default_evidence_level":"A","staleness_after_days":30,"notes":"Official endpoint catalog distinguishes Gemini 3.7, 3.6, 3.5 Flash and Flash-Lite; latest aliases may change and are not used as release identities here."},{"id":"src-xai-grok45","label":"SpaceXAI · Introducing Grok 4.5","kind":"primary","publisher":"xAI","url":"https://x.ai/news/grok-4-5","default_evidence_level":"B","staleness_after_days":180,"notes":"Official 2026-07-16 release announcement; effort, search and partner harness presentations remain protocol variants."},{"id":"src-xai-grok46","label":"SpaceXAI · Introducing Grok 4.6","kind":"primary","publisher":"xAI","url":"https://x.ai/news/grok-4-6","default_evidence_level":"B","staleness_after_days":120,"notes":"Official 2026-08-12 release announcement; benchmark rows must retain reasoning effort and agent harness."},{"id":"src-glm5v-turbo","label":"Z.ai · GLM-5V-Turbo documentation","kind":"documentation","publisher":"Z.ai","url":"https://docs.z.ai/guides/vlm/glm-5v-turbo","default_evidence_level":"A","staleness_after_days":90,"notes":"Official multimodal coding model guide; documents the glm-5v-turbo endpoint, modalities, 200K context and 128K maximum output."},{"id":"src-eigenigma-ikp-frontier-2026-07-31","label":"Eigenigma · Claude 5 / GPT-5.6 IKP capacity analysis","kind":"independent-analysis","publisher":"Eigenigma","url":"https://eigenigma.io/en/articles/estimating-parameter-counts-of-claude-5-and-gpt-5-6/","default_evidence_level":"C","staleness_after_days":180,"notes":"Reproducible black-box IKP analysis with downloadable per-question data. Values are IKP-effective knowledge-capacity equivalents with wide 90% intervals, not literal model weight counts; never use them as canonical params_total or params_active."},{"id":"src-openai-gpt55-spec","label":"OpenAI API · GPT-5.5 specification","kind":"documentation","publisher":"OpenAI","url":"https://developers.openai.com/api/docs/models/gpt-5.5","default_evidence_level":"A","staleness_after_days":30},{"id":"src-openai-gpt55-pro-spec","label":"OpenAI API · GPT-5.5 Pro specification","kind":"documentation","publisher":"OpenAI","url":"https://developers.openai.com/api/docs/models/gpt-5.5-pro","default_evidence_level":"A","staleness_after_days":30},{"id":"src-openai-gpt54-spec","label":"OpenAI API · GPT-5.4 specification","kind":"documentation","publisher":"OpenAI","url":"https://developers.openai.com/api/docs/models/gpt-5.4","default_evidence_level":"A","staleness_after_days":30},{"id":"src-openai-gpt54-mini-spec","label":"OpenAI API · GPT-5.4 mini specification","kind":"documentation","publisher":"OpenAI","url":"https://developers.openai.com/api/docs/models/gpt-5.4-mini","default_evidence_level":"A","staleness_after_days":30},{"id":"src-openai-gpt54-pro-spec","label":"OpenAI API · GPT-5.4 Pro specification","kind":"documentation","publisher":"OpenAI","url":"https://developers.openai.com/api/docs/models/gpt-5.4-pro","default_evidence_level":"A","staleness_after_days":30},{"id":"src-openai-gpt54-nano-spec","label":"OpenAI API · GPT-5.4 nano specification","kind":"documentation","publisher":"OpenAI","url":"https://developers.openai.com/api/docs/models/gpt-5.4-nano","default_evidence_level":"A","staleness_after_days":30},{"id":"src-openai-gpt53-codex-spec","label":"OpenAI API · GPT-5.3 Codex specification","kind":"documentation","publisher":"OpenAI","url":"https://developers.openai.com/api/docs/models/gpt-5.3-codex","default_evidence_level":"A","staleness_after_days":30},{"id":"src-anthropic-models-overview","label":"Anthropic · Claude model specification table","kind":"documentation","publisher":"Anthropic","url":"https://platform.claude.com/docs/en/about-claude/models/overview","default_evidence_level":"A","staleness_after_days":30},{"id":"src-qwen38-max-spec","label":"Qwen Cloud · Qwen3.8 Max specification","kind":"documentation","publisher":"Qwen","url":"https://docs.qwencloud.com/developer-guides/getting-started/latest-model","default_evidence_level":"A","staleness_after_days":30},{"id":"src-kimi-k2-hf","label":"Moonshot AI · Kimi K2 base model card","kind":"primary","publisher":"Moonshot AI","url":"https://huggingface.co/moonshotai/Kimi-K2-Instruct","default_evidence_level":"A","staleness_after_days":180,"notes":"Base-lineage evidence only for K2.7 coding endpoints; never promoted to exact K2.7 canonical parameter counts."},{"id":"src-qwen38-flash-hf","label":"Qwen · Qwen3.8 Flash-Next model card","kind":"primary","publisher":"Qwen","url":"https://huggingface.co/Qwen/Qwen3.8-Flash-Next","default_evidence_level":"A","staleness_after_days":90},{"id":"src-qwen38-27b-hf","label":"Qwen · Qwen3.8 27B model card","kind":"primary","publisher":"Qwen","url":"https://huggingface.co/Qwen/Qwen3.8-27B","default_evidence_level":"A","staleness_after_days":90},{"id":"src-qwen38-24t-hf","label":"Qwen · Qwen3.8 2.4T A95B model card","kind":"primary","publisher":"Qwen","url":"https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B","default_evidence_level":"A","staleness_after_days":90},{"id":"src-glm53-flash-hf","label":"Z.ai · GLM-5.3 Flash model card","kind":"primary","publisher":"Z.ai","url":"https://huggingface.co/zai-org/GLM-5.3-Flash","default_evidence_level":"A","staleness_after_days":90},{"id":"src-minimax-m3-hf","label":"MiniMax · MiniMax M3 model card","kind":"primary","publisher":"MiniMax","url":"https://huggingface.co/MiniMaxAI/MiniMax-M3","default_evidence_level":"A","staleness_after_days":90},{"id":"src-mimo-v25-hf","label":"Xiaomi MiMo · MiMo-V2.5 model card","kind":"primary","publisher":"Xiaomi MiMo","url":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5","default_evidence_level":"A","staleness_after_days":90},{"id":"src-google-gemini31-spec","label":"Google AI · Gemini 3.1 Pro endpoint specification","kind":"documentation","publisher":"Google DeepMind","url":"https://ai.google.dev/gemini-api/docs/models/gemini-3.1-pro-preview","default_evidence_level":"A","staleness_after_days":30},{"id":"src-google-gemini37-spec","label":"Google AI · Gemini 3.7 Flash endpoint specification","kind":"documentation","publisher":"Google DeepMind","url":"https://ai.google.dev/gemini-api/docs/models/gemini-3.7-flash","default_evidence_level":"A","staleness_after_days":30},{"id":"src-xai-grok46-spec","label":"xAI API · Grok 4.6 specification","kind":"documentation","publisher":"xAI","url":"https://docs.x.ai/developers/grok-4-6","default_evidence_level":"A","staleness_after_days":30}],"catalogModels":[{"id":"openai/gpt-5.6-sol@2026-07-09","canonicalId":"openai/gpt-5.6-sol@2026-07-09","familyId":"openai/gpt-5.6","name":"GPT-5.6 Sol","provider":"OpenAI","mark":"O","release":"2026-07-09","releaseDate":"2026-07-09","access":"api","status":"active","availability":"api","aliases":["gpt-5.6-sol","gpt-5.6"],"tags":["旗舰","推理","多模态","Agent"],"summary":"OpenAI GPT-5.6 reasoning tier；API alias gpt-5.6 路由到该 release。","modalities":["text","image"],"contextWindow":1050000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"tier":"sol","preview":false},"profile":{"positioning":"GPT-5.6 旗舰 reasoning tier，面向复杂专业工作。","capabilities":["复杂推理","代码","工具调用","电脑使用","视觉输入","多语言"],"best_for":["复杂研究与长程 agent","高质量代码与调试","需要最高准确度的专业任务"],"endpoint_ids":["gpt-5.6-sol","gpt-5.6"],"availability":["api","chat","codex"],"reasoning_modes":["none","low","medium","high","xhigh","max"],"architecture":null,"context_note":"API catalog 记录 1.05M context、128K max output；具体产品 surface 可能不同。","parameter_note":"官方未披露实际权重数；下方第三方估计衡量的是 IKP-effective 知识容量，不是 total / active 参数。","parameter_estimates":[{"kind":"ikp_effective","label":"IKP-effective knowledge capacity","point_b":1930,"range_b":[559,6700],"interval":"90%","basis":"attempted","confidence":"method-dependent","source_id":"src-eigenigma-ikp-frontier-2026-07-31","as_of":"2026-07-31","note":"第三方黑盒容量估计；表示与约 1.93T 平均开放模型相当的长尾知识容量，不能表述为该模型拥有 1.93T 权重。"}],"knowledge_cutoff":"2026-02-16","max_output_tokens":128000,"license":null,"caveats":["Sol/Terra/Luna 是能力与成本 tier，不应当作相同 endpoint 的平均分。","公开成绩必须同时记录 reasoning effort、工具和 harness。"],"fact_source_ids":["src-openai-models","src-openai-gpt56","src-eigenigma-ikp-frontier-2026-07-31"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-openai-gpt56"],"scores":{"gpqa-diamond":{"value":94.6,"raw_value":"94.6%","setting":"official GPT-5.6 evaluation table; no tools","sourceId":"src-openai-gpt56","sourceUrl":"https://openai.com/index/gpt-5-6/","sourceLabel":"OpenAI · GPT-5.6 launch and evaluations","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-07-09","published_at":"2026-07-09","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":0,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"official GPT-5.6 evaluation table; no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"obs-gpt56-sol-gpqa-20260827","notes":"Provider-reported value; exact reasoning budget and evaluator details remain in the linked launch table.","legacyMigrated":false},"mmmu-pro":{"value":83.0,"raw_value":"83.0%","setting":"MMMU-Pro; no tools","sourceId":"src-openai-gpt56","sourceUrl":"https://openai.com/index/gpt-5-6/","sourceLabel":"OpenAI · GPT-5.6 launch and evaluations","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-07-09","published_at":"2026-07-09","benchmark_version":"mmmu-pro@v1","version":"Pro","metric":"accuracy","unit":"%","protocol":{"shots":0,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"MMMU-Pro; no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"obs-gpt56-sol-mmmupro-no-tools-20260827","notes":"No-tools variant is preferred for the model atlas; the with-tools variant is retained as a separate run.","legacyMigrated":false}},"scoreCount":2,"catalogOnly":false,"systemRunCount":1},{"id":"openai/gpt-5.6-terra@2026-07-09","canonicalId":"openai/gpt-5.6-terra@2026-07-09","familyId":"openai/gpt-5.6","name":"GPT-5.6 Terra","provider":"OpenAI","mark":"O","release":"2026-07-09","releaseDate":"2026-07-09","access":"api","status":"active","availability":"api","aliases":["gpt-5.6-terra"],"tags":["通用","推理","多模态"],"summary":"GPT-5.6 balanced/production tier。","modalities":["text","image"],"contextWindow":1050000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"tier":"terra","preview":false},"profile":{"positioning":"GPT-5.6 balanced/production tier，在能力与成本之间取平衡。","capabilities":["推理","代码","工具调用","电脑使用","视觉输入","多语言"],"best_for":["通用生产工作流","中高难度代码和知识工作","需要稳定成本的 agent"],"endpoint_ids":["gpt-5.6-terra"],"availability":["api","chat","codex"],"reasoning_modes":["none","low","medium","high","xhigh","max"],"architecture":null,"context_note":"API catalog 记录 1.05M context、128K max output；具体产品 surface 可能不同。","parameter_note":"官方未披露实际权重数；下方第三方估计衡量的是 IKP-effective 知识容量，不是 total / active 参数。","parameter_estimates":[{"kind":"ikp_effective","label":"IKP-effective knowledge capacity","point_b":702,"range_b":[204,2400],"interval":"90%","basis":"attempted","confidence":"method-dependent","source_id":"src-eigenigma-ikp-frontier-2026-07-31","as_of":"2026-07-31","note":"第三方黑盒容量估计；表示与约 702B 平均开放模型相当的长尾知识容量，不能表述为该模型拥有 702B 权重。"}],"knowledge_cutoff":"2026-02-16","max_output_tokens":128000,"license":null,"caveats":["与 Sol/Luna 的成本和 effort 设置不同，跨 tier 分数需保留 endpoint。"],"fact_source_ids":["src-openai-models","src-openai-gpt56","src-eigenigma-ikp-frontier-2026-07-31"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-openai-gpt56"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.6-luna@2026-07-09","canonicalId":"openai/gpt-5.6-luna@2026-07-09","familyId":"openai/gpt-5.6","name":"GPT-5.6 Luna","provider":"OpenAI","mark":"O","release":"2026-07-09","releaseDate":"2026-07-09","access":"api","status":"active","availability":"api","aliases":["gpt-5.6-luna"],"tags":["快速","效率","多模态"],"summary":"GPT-5.6 lower-latency tier。","modalities":["text","image"],"contextWindow":1050000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"tier":"luna","preview":false},"profile":{"positioning":"GPT-5.6 的低延迟、低成本 tier，面向高吞吐工作负载。","capabilities":["快速推理","代码","工具调用","电脑使用","视觉输入","多语言"],"best_for":["高频 API 调用","交互式产品和路由","对延迟敏感的轻量 agent"],"endpoint_ids":["gpt-5.6-luna"],"availability":["api","chat","codex"],"reasoning_modes":["none","low","medium","high","xhigh","max"],"architecture":null,"context_note":"API catalog 记录 1.05M context、128K max output；具体产品 surface 可能不同。","parameter_note":"官方未披露实际权重数；下方第三方估计衡量的是 IKP-effective 知识容量，不是 total / active 参数。","parameter_estimates":[{"kind":"ikp_effective","label":"IKP-effective knowledge capacity","point_b":744,"range_b":[216,2600],"interval":"90%","basis":"attempted","confidence":"method-dependent","source_id":"src-eigenigma-ikp-frontier-2026-07-31","as_of":"2026-07-31","note":"第三方黑盒容量估计；表示与约 744B 平均开放模型相当的长尾知识容量，不能表述为该模型拥有 744B 权重。"}],"knowledge_cutoff":"2026-02-16","max_output_tokens":128000,"license":null,"caveats":["低价/低延迟定位不等于与 Sol 使用同一计算预算；比较时应标注 tier。"],"fact_source_ids":["src-openai-models","src-openai-gpt56","src-eigenigma-ikp-frontier-2026-07-31"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-openai-gpt56"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.5@current","canonicalId":"openai/gpt-5.5@current","familyId":"openai/gpt-5.5","name":"GPT-5.5","provider":"OpenAI","mark":"O","release":"2026-04-23","releaseDate":"2026-04-23","access":"api","status":"active","availability":"api","aliases":["gpt-5.5","gpt-5-5"],"tags":["推理","代码","Agent"],"summary":"上一代 OpenAI frontier baseline。","modalities":["text","image"],"contextWindow":1050000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"OpenAI 的通用 frontier baseline，覆盖知识工作、代码和工具型工作流。","capabilities":["推理","代码","知识工作","工具调用","电脑使用","视觉输入"],"best_for":["长程知识工作","代码 agent","专业文档与数据任务"],"endpoint_ids":["gpt-5.5"],"availability":["chat","codex","api"],"reasoning_modes":["thinking","pro"],"architecture":null,"context_note":"官方 endpoint 规格记录 1,050,000-token context；具体配额仍可能随 rollout 变化。","parameter_note":"参数量未公开。","knowledge_cutoff":"2025-12-01","max_output_tokens":128000,"license":null,"caveats":["GPT-5.5 与 GPT-5.5 Pro 是产品/计算 tier；不要把 Pro 当作独立权重而忽略 endpoint。","发布页中的内部评测和公开 benchmark 需区分。"],"fact_source_ids":["src-openai-gpt55","src-openai-gpt55-spec"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-gpt55","src-openai-gpt55-spec"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.5-pro@current","canonicalId":"openai/gpt-5.5-pro@current","familyId":"openai/gpt-5.5","name":"GPT-5.5 Pro","provider":"OpenAI","mark":"O","release":null,"releaseDate":null,"access":"api","status":"active","availability":"api","aliases":["gpt-5.5-pro"],"tags":["推理","代码"],"summary":"GPT-5.5 higher-compute product tier；作为独立 endpoint 记录。","modalities":["text","image"],"contextWindow":1050000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"tier":"pro","preview":false},"profile":{"positioning":"GPT-5.5 的高计算、最高质量产品 tier，面向最难的专业任务。","capabilities":["深度推理","代码","知识工作","工具调用","视觉输入"],"best_for":["高准确度研究","复杂分析与审阅","高价值低吞吐任务"],"endpoint_ids":["gpt-5.5-pro"],"availability":["chat","codex","api"],"reasoning_modes":["pro"],"architecture":null,"context_note":"官方 endpoint 规格记录 1,050,000-token context；API rollout 和价格可能变化。","parameter_note":"参数量未公开。","knowledge_cutoff":"2025-12-01","max_output_tokens":128000,"license":null,"caveats":["这是高计算 serving tier；不能与默认 GPT-5.5 的结果直接混排。","模型页当前将其作为独立 endpoint 记录，以保留可追溯性。"],"fact_source_ids":["src-openai-gpt55","src-openai-gpt55-pro-spec"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-gpt55","src-openai-gpt55-pro-spec"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.4@current","canonicalId":"openai/gpt-5.4@current","familyId":"openai/gpt-5.4","name":"GPT-5.4","provider":"OpenAI","mark":"O","release":"2026-03-05","releaseDate":"2026-03-05","access":"api","status":"active","availability":"api","aliases":["gpt-5.4","gpt-5-4"],"tags":["通用","推理"],"summary":"","modalities":["text","image"],"contextWindow":1050000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"面向专业工作的主线 reasoning 模型，强调代码、工具和原生 computer use。","capabilities":["推理","代码","电脑使用","视觉","工具搜索","专业知识工作"],"best_for":["浏览器/桌面 agent","文档、表格和演示文稿","长程代码任务"],"endpoint_ids":["gpt-5.4","gpt-5.4-pro"],"availability":["api","chat","codex"],"reasoning_modes":["none","low","medium","high","xhigh"],"architecture":null,"context_note":"当前官方 API endpoint 规格记录 1,050,000-token context；历史产品面与 Codex 配置不能直接代入其他 endpoint。","parameter_note":"参数量未公开。","knowledge_cutoff":"2025-08-31","max_output_tokens":null,"license":null,"caveats":["computer-use 成绩高度依赖工具、截图/DOM 设置和 harness。","Pro 是独立 serving tier，需单独记录。"],"fact_source_ids":["src-openai-models","src-openai-gpt54-spec"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-openai-gpt54-spec"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.4-mini@current","canonicalId":"openai/gpt-5.4-mini@current","familyId":"openai/gpt-5.4","name":"GPT-5.4 mini","provider":"OpenAI","mark":"O","release":"2026-03-17","releaseDate":"2026-03-17","access":"api","status":"active","availability":"api","aliases":["gpt-5.4-mini"],"tags":["小模型","效率"],"summary":"","modalities":["text","image"],"contextWindow":400000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"tier":"mini","preview":false},"profile":{"positioning":"GPT-5.4 的效率/小型 tier，优先考虑吞吐、成本和交互延迟。","capabilities":["轻量推理","代码","工具调用","视觉输入"],"best_for":["高吞吐应用","路由后的中等难度任务","成本敏感的代码辅助"],"endpoint_ids":["gpt-5.4-mini"],"availability":["api"],"reasoning_modes":["none","low","medium","high"],"architecture":null,"context_note":"官方 endpoint 规格记录 400,000-token context。","parameter_note":"参数量未公开；mini 是产品 tier，不等于公开权重规模。","knowledge_cutoff":"2025-08-31","max_output_tokens":null,"license":null,"caveats":["不要用 GPT-5.4 主模型的 benchmark 分数代填 mini。"],"fact_source_ids":["src-openai-models","src-openai-gpt54-mini-spec"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-openai-gpt54-mini-spec"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5@2025-08-07","canonicalId":"openai/gpt-5@2025-08-07","familyId":"openai/gpt-5","name":"GPT-5","provider":"OpenAI","mark":"O","release":"2025-08-07","releaseDate":"2025-08-07","access":"api","status":"previous","availability":"api","aliases":["gpt-5"],"tags":["上一代","推理","工具"],"summary":"Legacy MVP baseline; GPT-5 pro values remain identified in observation notes.","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GPT-5 代际 baseline，保留用于历史纵向比较。","capabilities":["推理","代码","工具调用","视觉输入"],"best_for":["与 GPT-5.4/5.5 的代际对照","历史 benchmark 时间线"],"endpoint_ids":["gpt-5"],"availability":["api","chat"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"本目录未从当前公开资料统一核验 context，保留未注明。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["旧 release 的榜单值可能使用不同 endpoint、prompt 或工具；必须查看 observation 详情。"],"fact_source_ids":["openai-gpt5"],"last_checked":"2026-08-27"},"sourceIds":["openai-gpt5"],"scores":{"gpqa-diamond":{"value":88.4,"raw_value":"88.4%","setting":"GPT-5 pro · no tools","sourceId":"openai-gpt5","sourceUrl":"https://openai.com/index/introducing-gpt-5/","sourceLabel":"OpenAI · Introducing GPT-5","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-08-07","published_at":"2025-08-07","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"GPT-5 pro · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gpt-5-gpqa-diamond","notes":"该值是 GPT-5 pro 变体，不应与普通 GPT-5 endpoint 混为同一 release。","legacyMigrated":true},"aime-2025":{"value":94.6,"raw_value":"94.6%","setting":"no tools · high","sourceId":"openai-gpt5","sourceUrl":"https://openai.com/index/introducing-gpt-5/","sourceLabel":"OpenAI · Introducing GPT-5","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-08-07","published_at":"2025-08-07","benchmark_version":"aime-2025@v1","version":"2025","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools · high"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gpt-5-aime-2025","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"mmmu":{"value":84.2,"raw_value":"84.2%","setting":"MMMU · no tools","sourceId":"openai-gpt5","sourceUrl":"https://openai.com/index/introducing-gpt-5/","sourceLabel":"OpenAI · Introducing GPT-5","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-08-07","published_at":"2025-08-07","benchmark_version":"mmmu@v1","version":"original","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gpt-5-mmmu","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":3,"catalogOnly":false,"systemRunCount":1},{"id":"anthropic/claude-opus-5@2026-07-24","canonicalId":"anthropic/claude-opus-5@2026-07-24","familyId":"anthropic/claude-opus","name":"Claude Opus 5","provider":"Anthropic","mark":"A","release":"2026-07-24","releaseDate":"2026-07-24","access":"api","status":"active","availability":"api","aliases":["claude-opus-5"],"tags":["旗舰","推理","代码","Agent"],"summary":"Anthropic current Opus flagship。","modalities":["text","image"],"contextWindow":1000000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Anthropic 当前 Opus 旗舰，面向复杂推理、代码和 agent 工作。","capabilities":["深度推理","代码","视觉输入","长程 agent","工具调用"],"best_for":["复杂研究与编程","高难度 agent workflow","长文档和多步任务"],"endpoint_ids":["claude-opus-5"],"availability":["api","hosted"],"reasoning_modes":["extended thinking"],"architecture":null,"context_note":"目录记录 1M context；Fast mode 约为默认速度的 2.5 倍，属于 serving mode。","parameter_note":"官方未披露实际权重数；下方第三方估计是经拒答校正的 IKP-effective 知识容量，不是 total / active 参数。","parameter_estimates":[{"kind":"ikp_effective","label":"IKP-effective knowledge capacity","point_b":2100,"range_b":[609,7200],"interval":"90%","basis":"sol-imputed","confidence":"method-dependent","source_id":"src-eigenigma-ikp-frontier-2026-07-31","as_of":"2026-07-31","note":"第三方黑盒容量估计；作者用 GPT-5.6 Sol 对拒答样本做 imputation。区间尚未计入 imputation 的额外不确定性，不能当作实际权重数。"}],"knowledge_cutoff":"2026-05","max_output_tokens":128000,"license":null,"caveats":["Fast mode 改变延迟/价格，不应与默认模式合并为单一分数。","安全设置和 tool scaffold 会显著影响 agent benchmark。"],"fact_source_ids":["src-anthropic-opus5","src-anthropic-system-cards","src-anthropic-models-overview","src-eigenigma-ikp-frontier-2026-07-31"],"last_checked":"2026-08-27"},"sourceIds":["src-anthropic-opus5","src-anthropic-system-cards"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"anthropic/claude-opus@4","canonicalId":"anthropic/claude-opus@4","familyId":"anthropic/claude-opus","name":"Claude Opus 4","provider":"Anthropic","mark":"A","release":"2025-05-22","releaseDate":"2025-05-22","access":"api","status":"previous","availability":"api","aliases":["claude-opus-4"],"tags":["上一代","代码","推理","Agent"],"summary":"Legacy MVP baseline; extended thinking and Claude Code scaffold are recorded per observation.","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Claude Opus 4 历史旗舰，作为 Opus 系列上一代对照。","capabilities":["推理","代码","视觉输入","Agent","extended thinking"],"best_for":["代际比较","历史 Claude Code / agent 结果"],"endpoint_ids":["claude-opus-4"],"availability":["api","hosted"],"reasoning_modes":["extended thinking"],"architecture":null,"context_note":"本目录未统一核验当前 context/配额。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["历史结果经常绑定 Claude Code scaffold；不能当作裸模型结果。"],"fact_source_ids":["anthropic-claude4"],"last_checked":"2026-08-27"},"sourceIds":["anthropic-claude4"],"scores":{"gpqa-diamond":{"value":83.3,"raw_value":"83.3%","setting":"extended thinking · 64K","sourceId":"anthropic-claude4","sourceUrl":"https://www.anthropic.com/news/claude-4","sourceLabel":"Anthropic · Introducing Claude 4","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-22","published_at":"2025-05-22","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"extended thinking · 64K"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-claude-opus-4-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"aime-2025":{"value":75.5,"raw_value":"75.5%","setting":"single attempt · extended thinking","sourceId":"anthropic-claude4","sourceUrl":"https://www.anthropic.com/news/claude-4","sourceLabel":"Anthropic · Introducing Claude 4","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-22","published_at":"2025-05-22","benchmark_version":"aime-2025@v1","version":"2025","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"single attempt · extended thinking"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-claude-opus-4-aime-2025","notes":"发布表另列出并行 test-time compute 的 90.0%，此处保留单次值。","legacyMigrated":true},"mmmu":{"value":76.5,"raw_value":"76.5%","setting":"MMMU validation · extended thinking","sourceId":"anthropic-claude4","sourceUrl":"https://www.anthropic.com/news/claude-4","sourceLabel":"Anthropic · Introducing Claude 4","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-22","published_at":"2025-05-22","benchmark_version":"mmmu@v1","version":"original","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU validation · extended thinking"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-claude-opus-4-mmmu","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":3,"catalogOnly":false,"systemRunCount":2},{"id":"anthropic/claude-fable-5@2026-06-09","canonicalId":"anthropic/claude-fable-5@2026-06-09","familyId":"anthropic/claude-fable","name":"Claude Fable 5","provider":"Anthropic","mark":"A","release":"2026-06-09","releaseDate":"2026-06-09","access":"restricted","status":"restricted","availability":"restricted","aliases":["claude-fable-5","anthropic-claude-fable-5"],"tags":["前沿","推理","代码","restricted"],"summary":"Trusted/restricted Fable endpoint；报告时保留 fallback policy。","modalities":["text","image"],"contextWindow":1000000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"fallback_policy":"may fallback to Opus 4.8"},"profile":{"positioning":"受限访问的 Fable 5 endpoint，面向高难度推理与代码工作。","capabilities":["推理","代码","视觉输入","Agent","长上下文"],"best_for":["受控研究访问","需要 frontier reasoning 的 agent 任务"],"endpoint_ids":["claude-fable-5"],"availability":["restricted","hosted"],"reasoning_modes":["extended thinking"],"architecture":null,"context_note":"目录记录 1M context；具体 trusted-access 配额未统一公开。","parameter_note":"官方未披露实际权重数；下方第三方估计是经拒答校正的 IKP-effective 知识容量，不是 total / active 参数。","parameter_estimates":[{"kind":"ikp_effective","label":"IKP-effective knowledge capacity","point_b":2100,"range_b":[609,7200],"interval":"90%","basis":"sol-imputed","confidence":"method-dependent","source_id":"src-eigenigma-ikp-frontier-2026-07-31","as_of":"2026-07-31","note":"第三方黑盒容量估计；作者用 GPT-5.6 Sol 对拒答样本做 imputation。区间尚未计入 imputation 的额外不确定性，不能当作实际权重数。"}],"knowledge_cutoff":"2026-01","max_output_tokens":128000,"license":null,"caveats":["安全分类器可能 fallback 到 Opus 4.8；评测必须记录是否发生 fallback。","restricted endpoint 不应与公开 API 结果直接混排。"],"fact_source_ids":["src-anthropic-fable5","src-anthropic-system-cards","src-anthropic-models-overview","src-eigenigma-ikp-frontier-2026-07-31"],"last_checked":"2026-08-27"},"sourceIds":["src-anthropic-fable5","src-anthropic-system-cards"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"anthropic/claude-mythos-5@2026-06-09","canonicalId":"anthropic/claude-mythos-5@2026-06-09","familyId":"anthropic/claude-mythos","name":"Claude Mythos 5","provider":"Anthropic","mark":"A","release":"2026-06-09","releaseDate":"2026-06-09","access":"restricted","status":"restricted","availability":"restricted","aliases":["claude-mythos-5"],"tags":["前沿","推理","restricted"],"summary":"与 Fable 5 同 base model 的受限研究访问变体。","modalities":["text","image"],"contextWindow":1000000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"safeguards":"trusted access only"},"profile":{"positioning":"与 Fable 5 同基础模型谱系、但面向 trusted access 的 Mythos 5 变体。","capabilities":["深度推理","代码","视觉输入","Agent","长上下文"],"best_for":["受控安全研究","需要更高能力上限的专业任务"],"endpoint_ids":["claude-mythos-5"],"availability":["restricted"],"reasoning_modes":["extended thinking"],"architecture":null,"context_note":"目录记录 1M context；trusted-access 限制和配额需以 provider 页面为准。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["不能把 Mythos 与 Fable 的安全策略、fallback 或可访问性视为相同。","公开比较应保留 restricted 标签。"],"fact_source_ids":["src-anthropic-fable5","src-anthropic-system-cards"],"last_checked":"2026-08-27"},"sourceIds":["src-anthropic-fable5","src-anthropic-system-cards"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"anthropic/claude-sonnet-5@2026-06-30","canonicalId":"anthropic/claude-sonnet-5@2026-06-30","familyId":"anthropic/claude-sonnet","name":"Claude Sonnet 5","provider":"Anthropic","mark":"A","release":"2026-06-30","releaseDate":"2026-06-30","access":"api","status":"active","availability":"api","aliases":["claude-sonnet-5"],"tags":["通用","推理","代码","Agent"],"summary":"Sonnet 5 current production model with adaptive thinking。","modalities":["text","image"],"contextWindow":1000000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"adaptive_thinking":true,"preview":false},"profile":{"positioning":"Claude Sonnet 5 production tier，在能力、速度和成本之间平衡。","capabilities":["推理","代码","视觉输入","Agent","adaptive thinking"],"best_for":["日常生产 agent","代码协作","多模态知识工作"],"endpoint_ids":["claude-sonnet-5"],"availability":["api","hosted"],"reasoning_modes":["adaptive thinking"],"architecture":null,"context_note":"目录记录 1M context；不同 API/hosted surface 的输出上限需单独核验。","parameter_note":"官方未披露实际权重数；下方第三方估计是经拒答校正的 IKP-effective 知识容量，不是 total / active 参数。","parameter_estimates":[{"kind":"ikp_effective","label":"IKP-effective knowledge capacity","point_b":972,"range_b":[282,3400],"interval":"90%","basis":"sol-imputed","confidence":"method-dependent","source_id":"src-eigenigma-ikp-frontier-2026-07-31","as_of":"2026-07-31","note":"第三方黑盒容量估计；作者用 GPT-5.6 Sol 对拒答样本做 imputation。区间尚未计入 imputation 的额外不确定性，不能当作实际权重数。"}],"knowledge_cutoff":"2026-01","max_output_tokens":128000,"license":null,"caveats":["adaptive thinking 是运行模式，不是新的权重 release。","工具和 prompt template 会影响 agent 结果。"],"fact_source_ids":["src-anthropic-sonnet5","src-anthropic-system-cards","src-anthropic-models-overview","src-eigenigma-ikp-frontier-2026-07-31"],"last_checked":"2026-08-27"},"sourceIds":["src-anthropic-sonnet5","src-anthropic-system-cards"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"anthropic/claude-opus-4.8@current","canonicalId":"anthropic/claude-opus-4.8@current","familyId":"anthropic/claude-opus","name":"Claude Opus 4.8","provider":"Anthropic","mark":"A","release":null,"releaseDate":null,"access":"api","status":"active","availability":"api","aliases":["claude-opus-4-8"],"tags":["上一代","代码","推理"],"summary":"","modalities":["text","image"],"contextWindow":1000000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Opus 4.8 serving release，作为 Opus 5 的近代对照和 fallback 目标。","capabilities":["推理","代码","视觉输入","Agent"],"best_for":["Opus 5 fallback 对照","历史/近代 agent benchmark"],"endpoint_ids":["claude-opus-4.8"],"availability":["api","hosted"],"reasoning_modes":["extended thinking"],"architecture":null,"context_note":"目录记录 1M context；具体产品限制需按 endpoint 复核。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["作为 fallback 时，系统实际 subject 可能是 Fable/Opus + fallback，而非裸 Opus 4.8。"],"fact_source_ids":["src-anthropic-system-cards"],"last_checked":"2026-08-27"},"sourceIds":["src-anthropic-system-cards"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"anthropic/claude-opus-4.7@current","canonicalId":"anthropic/claude-opus-4.7@current","familyId":"anthropic/claude-opus","name":"Claude Opus 4.7","provider":"Anthropic","mark":"A","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["claude-opus-4-7"],"tags":["历史","代码"],"summary":"","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Opus 系列上一代代码/推理 baseline。","capabilities":["推理","代码","视觉输入","Agent"],"best_for":["与 Opus 5 的代际比较","历史代码 agent 结果"],"endpoint_ids":["claude-opus-4.7"],"availability":["api","hosted"],"reasoning_modes":["extended thinking"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["不同 system card / Claude Code scaffold 的结果不可直接合并。"],"fact_source_ids":["src-anthropic-system-cards"],"last_checked":"2026-08-27"},"sourceIds":["src-anthropic-system-cards"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"anthropic/claude-sonnet-4.6@current","canonicalId":"anthropic/claude-sonnet-4.6@current","familyId":"anthropic/claude-sonnet","name":"Claude Sonnet 4.6","provider":"Anthropic","mark":"A","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["claude-sonnet-4-6"],"tags":["上一代","效率","代码"],"summary":"","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Sonnet 5 的上一代效率与代码 baseline。","capabilities":["推理","代码","视觉输入","Agent"],"best_for":["Sonnet 代际比较","成本/质量平衡的历史基线"],"endpoint_ids":["claude-sonnet-4.6"],"availability":["api","hosted"],"reasoning_modes":["extended thinking"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["上一代模型仍可能在 provider/aggregator 上路由到不同 snapshot。"],"fact_source_ids":["src-anthropic-system-cards"],"last_checked":"2026-08-27"},"sourceIds":["src-anthropic-system-cards"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.8-max@2026-08-02","canonicalId":"qwen/qwen3.8-max@2026-08-02","familyId":"qwen/qwen3.8","name":"Qwen3.8 Max","provider":"Qwen","mark":"Q","release":"2026-08-02","releaseDate":"2026-08-02","access":"api","status":"active","availability":"api","aliases":["qwen3.8-max","qwen-3-8-max"],"tags":["旗舰","托管服务","推理","代码"],"summary":"Qwen3.8 Max 托管旗舰 endpoint；官方规格披露 1M context 与 2.4T MoE 总参数，95B active 仅作为底座 checkpoint 证据展示。","modalities":["text","image"],"contextWindow":1000000,"paramsTotal":2400000000000,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.8 Max 托管旗舰 endpoint；本目录将它与 Qwen3.8 2.4T A95B open-weight checkpoint 分开登记。","capabilities":["复杂推理","代码","视觉","多语言"],"best_for":["中文/多语言知识工作","复杂代码和推理","托管多模态工作"],"endpoint_ids":["qwen3.8-max","qwen-3-8-max"],"availability":["api","hosted"],"reasoning_modes":["thinking"],"architecture":"Mixture-of-Experts（托管 endpoint 的高层官方描述）","context_note":"Qwen Cloud 逐型号规格记录 1,000,000-token hosted context。","parameter_note":"官方托管规格披露 2.4T 总参数；95B active 只在底座 Qwen3.8-2.4T-A95B checkpoint 上直接披露，因此不写入 Max 的 canonical active 字段。","parameter_evidence":[{"kind":"base_checkpoint","label":"Qwen3.8-2.4T-A95B base checkpoint","total_b":2400,"active_b":95,"source_id":"src-qwen38-24t-hf","confidence":"vendor-published lineage","note":"托管 Max 明确基于 2.4T MoE；95B active 是底座 checkpoint 的数值，不代表对 hosted Max 的独立测量。"}],"knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["Max 服务名与 Qwen3.8 2.4T A95B open-weight checkpoint 不应自动视为同一权重或 endpoint。","当前来源只支持 family/discovery scope；架构、参数和 context 保持未知。"],"fact_source_ids":["src-qwen38-repo","src-qwen38-flash","src-qwen38-max-spec","src-qwen38-24t-hf"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen38-repo","src-qwen38-flash","src-qwen38-max-spec"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.8-flash-next@2026-08-26","canonicalId":"qwen/qwen3.8-flash-next@2026-08-26","familyId":"qwen/qwen3.8","name":"Qwen3.8 Flash-Next","provider":"Qwen","mark":"Q","release":"2026-08-26","releaseDate":"2026-08-26","access":"open_weights","status":"active","availability":"open_weights","aliases":["Qwen3.8-Flash-Next","qwen3.8-flash"],"tags":["Flash","效率","开放权重","代码"],"summary":"125B main / 6B active；另有 deterministic n-gram memory，不能当作 active parameters。","modalities":["text"],"contextWindow":262144,"paramsTotal":125000000000,"paramsActive":6000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false,"long_context_max":1010000},"profile":{"positioning":"Qwen3.8 Flash-Next 高吞吐 open-weight variant，强调低成本和长上下文。","capabilities":["推理","代码","长上下文","中文/多语言"],"best_for":["高吞吐代码任务","成本敏感 agent","长文档批处理"],"endpoint_ids":["Qwen/Qwen3.8-Flash-Next","qwen3.8-flash"],"availability":["open-weights","hosted-preview"],"reasoning_modes":["thinking"],"architecture":"Hybrid Gated DeltaNet + Qwen Sparse Attention + n-gram embeddings + MoE","context_note":"262,144 native；官方 model card 描述通过指定 scaling configuration 扩展到约 1.01M。","parameter_note":"125B main / 6B active；另有 51B n-gram embeddings 与 4B MTP 组件，不能混入 6B per-token active。","knowledge_cutoff":null,"max_output_tokens":null,"license":"qwen-community-1.0","caveats":["n-gram memory 是额外组件，不应加进模型 active 参数。","Flash-Next checkpoint 与 hosted qwen3.8-flash 的 serving 版本需分开记录。"],"fact_source_ids":["src-qwen38-flash","src-qwen38-repo","src-qwen38-flash-hf"],"last_checked":"2026-08-27"},"sourceIds":["src-qwen38-flash","src-qwen38-repo","src-qwen38-flash-hf"],"scores":{"gpqa-diamond":{"value":91.7,"raw_value":"91.7%","setting":"Qwen3.8-Flash-Next official table","sourceId":"src-qwen38-flash","sourceUrl":"https://qwen.ai/blog?id=qwen3.8-flash-next","sourceLabel":"Qwen · Qwen3.8-Flash-Next","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-08-26","published_at":"2026-08-26","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"Qwen3.8-Flash-Next official table"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"obs-qwen38-flash-gpqa-20260826","notes":"Provider-reported value; protocol details follow the linked Qwen release page.","legacyMigrated":false},"livecodebench":{"value":91.9,"raw_value":"91.9%","setting":"LiveCodeBench v6; Qwen official table","sourceId":"src-qwen38-flash","sourceUrl":"https://qwen.ai/blog?id=qwen3.8-flash-next","sourceLabel":"Qwen · Qwen3.8-Flash-Next","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-08-26","published_at":"2026-08-26","benchmark_version":"livecodebench@v6","version":"V6","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"LiveCodeBench v6; Qwen official table"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"obs-qwen38-flash-lcb-20260826","notes":"Provider-reported rolling benchmark value; window and evaluator should be checked before exact comparison.","legacyMigrated":false}},"scoreCount":2,"catalogOnly":false,"systemRunCount":1},{"id":"qwen/qwen3.8-27b@2026-08-14","canonicalId":"qwen/qwen3.8-27b@2026-08-14","familyId":"qwen/qwen3.8","name":"Qwen3.8 27B","provider":"Qwen","mark":"Q","release":"2026-08-14","releaseDate":"2026-08-14","access":"open_weights","status":"active","availability":"open_weights","aliases":["Qwen3.8-27B","qwen-3-8-27b"],"tags":["小模型","开放权重","推理","代码"],"summary":"","modalities":["text"],"contextWindow":262144,"paramsTotal":27000000000,"paramsActive":27000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.8 27B dense open-weight 模型，作为小模型/效率对照。","capabilities":["推理","代码","中文/多语言"],"best_for":["本地部署","小模型能力比较","成本受限的代码/文本任务"],"endpoint_ids":["Qwen/Qwen3.8-27B"],"availability":["open-weights"],"reasoning_modes":["thinking"],"architecture":"Dense Transformer","context_note":"原生 context 262,144；官方 model card 描述可通过指定 scaling configuration 扩展到约 1.01M。","parameter_note":"27B total / 27B active。","knowledge_cutoff":null,"max_output_tokens":null,"license":"Apache-2.0","caveats":["本地量化、推理框架和 batch size 会改变吞吐，不改变基础参数量。"],"fact_source_ids":["src-qwen38-27b","src-qwen38-repo","src-qwen38-27b-hf"],"last_checked":"2026-08-27"},"sourceIds":["src-qwen38-27b","src-qwen38-repo","src-qwen38-27b-hf"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.7-plus@current","canonicalId":"qwen/qwen3.7-plus@current","familyId":"qwen/qwen3.7","name":"Qwen3.7 Plus","provider":"Qwen","mark":"Q","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["qwen3.7-plus"],"tags":["上一代","托管服务","推理"],"summary":"Qwen3.7 Plus 托管 endpoint；参数规模、架构与 context 尚未从该 endpoint 的一手资料单独核验。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.7 Plus 上一代托管服务 endpoint，用于 Qwen 3.8/3.7 的服务代际对照。","capabilities":["推理","代码","视觉","中文/多语言","Agent"],"best_for":["Qwen 3.8/3.7 代际比较","托管中文/多模态工作"],"endpoint_ids":["qwen3.7-plus"],"availability":["api","hosted"],"reasoning_modes":["thinking"],"architecture":null,"context_note":"该托管 endpoint 的 context 上限尚未从逐型号的一手 endpoint 文档核验。","parameter_note":"该托管 endpoint 的参数量未公开；不得从名称相近但不同的 open-weight checkpoint 继承。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["服务 endpoint 与名称相近的公开权重不应自动视为同一 release。","架构、参数和 context 在逐型号一手资料确认前保持未知。"],"fact_source_ids":["src-qwen38-repo"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen38-repo"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.6-max@preview","canonicalId":"qwen/qwen3.6-max@preview","familyId":"qwen/qwen3.6","name":"Qwen3.6 Max Preview","provider":"Qwen","mark":"Q","release":"2026-04-18","releaseDate":"2026-04-18","access":"api","status":"preview","availability":"api","aliases":["qwen3.6-max-preview"],"tags":["上一代","preview","推理"],"summary":"","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":true},"profile":{"positioning":"Qwen3.6 Max Preview，作为正在迭代的前沿预览条目。","capabilities":["推理","代码","视觉","中文/多语言"],"best_for":["预览能力观察","Qwen 代际时间线"],"endpoint_ids":["qwen3.6-max-preview"],"availability":["hosted-preview","api"],"reasoning_modes":["thinking"],"architecture":null,"context_note":"当前公开页面未统一核验 context。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["Preview 结果可能随 checkpoint/服务路由变化，不进入严格稳定排名。"],"fact_source_ids":["src-qwen36"],"last_checked":"2026-08-27"},"sourceIds":["src-qwen36"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3-max@thinking","canonicalId":"qwen/qwen3-max@thinking","familyId":"qwen/qwen3-max","name":"Qwen3-Max-Thinking","provider":"Qwen","mark":"Q","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["qwen3-max-thinking","qwen3-max-2026-01-23"],"tags":["上一代","推理","代码","多语言"],"summary":"Legacy seed model；保留用于纵向比较。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3-Max-Thinking 历史 reasoning 服务 release。","capabilities":["深度推理","代码","工具调用","中文/多语言"],"best_for":["Qwen reasoning 时间线","复杂数学/代码历史对照"],"endpoint_ids":["qwen3-max-2026-01-23","qwen3-max-thinking"],"availability":["api","hosted"],"reasoning_modes":["thinking"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["Thinking 是运行模式/endpoint 标识，不应和 Qwen3-Max-Instruct 混合。"],"fact_source_ids":["qwen3-max"],"last_checked":"2026-08-27"},"sourceIds":["qwen3-max"],"scores":{"gpqa-diamond":{"value":87.4,"raw_value":"87.4%","setting":"官方横向表 · no custom scaffold","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-01-15","published_at":"2026-01-15","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"官方横向表 · no custom scaffold"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-qwen3-max-thinking-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"livecodebench":{"value":85.9,"raw_value":"85.9%","setting":"LiveCodeBench v6","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-01-15","published_at":"2026-01-15","benchmark_version":"livecodebench@v5","version":"V5","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"LiveCodeBench v6"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-qwen3-max-thinking-livecodebench","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"mmlu-pro":{"value":85.7,"raw_value":"85.7%","setting":"官方横向表","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-01-15","published_at":"2026-01-15","benchmark_version":"mmlu-pro@v1","version":"v1","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"官方横向表"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-qwen3-max-thinking-mmlu-pro","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"hle":{"value":30.2,"raw_value":"30.2%","setting":"text subset · no tools","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-01-15","published_at":"2026-01-15","benchmark_version":"hle@v1","version":"full","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"text subset · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-qwen3-max-thinking-hle","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":4,"catalogOnly":false,"systemRunCount":1},{"id":"qwen/qwen3-coder@current","canonicalId":"qwen/qwen3-coder@current","familyId":"qwen/qwen3-coder","name":"Qwen3-Coder 480B-A35B","provider":"Qwen","mark":"Q","release":null,"releaseDate":null,"access":"open_weights","status":"active","availability":"open_weights","aliases":["qwen3-coder","Qwen/Qwen3-Coder-480B-A35B-Instruct"],"tags":["代码","开放权重","专用"],"summary":"Qwen3-Coder 480B total / 35B active；原生 256K context，可扩展至 1M，扩展配置必须随 run 记录。","modalities":["text"],"contextWindow":262144,"paramsTotal":480000000000,"paramsActive":35000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false,"long_context_max":1000000},"profile":{"positioning":"Qwen3-Coder 480B-A35B 开放权重代码模型，面向仓库级软件工程与 agent 工作流。","capabilities":["代码生成","代码补全","仓库理解","工具调用"],"best_for":["本地代码助手","代码 agent","仓库级重构"],"endpoint_ids":["qwen3-coder","Qwen/Qwen3-Coder-480B-A35B-Instruct"],"availability":["open-weights"],"reasoning_modes":["thinking"],"architecture":"Mixture-of-Experts, 480B total / 35B active","context_note":"原生 256K（262,144 tokens）；模型卡给出可扩展至 1M 的配置，扩展方法与实际服务限制必须随 run 记录。","parameter_note":"480B total / 35B active。","knowledge_cutoff":null,"max_output_tokens":null,"license":"Apache-2.0","caveats":["专用 coding 模型与通用 Qwen release 不应仅按名称比较。","1M 是扩展配置而非原生 context；SWE/Terminal 类成绩必须保留 harness、工具和预算。"],"fact_source_ids":["src-qwen3-coder"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen3-coder"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"deepseek/v4-pro@2026-04-24","canonicalId":"deepseek/v4-pro@2026-04-24","familyId":"deepseek/v4","name":"DeepSeek V4 Pro","provider":"DeepSeek","mark":"D","release":"2026-04-24","releaseDate":"2026-04-24","access":"api","status":"active","availability":"api","aliases":["deepseek-v4-pro"],"tags":["旗舰","开放权重","推理","代码"],"summary":"","modalities":["text"],"contextWindow":1000000,"paramsTotal":1600000000000,"paramsActive":49000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"DeepSeek V4 Pro 旗舰 MoE release，面向推理、代码和长程任务。","capabilities":["深度推理","代码","长上下文","Agent"],"best_for":["复杂代码/数学","长上下文 agent","开放权重规模对比"],"endpoint_ids":["deepseek-v4-pro"],"availability":["api","hosted","open-weights"],"reasoning_modes":["reasoning"],"architecture":"MoE","context_note":"目录记录 1M context；API 与本地权重的限制要分开。","parameter_note":"1.6T total / 49B active。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["V4 preview/API release 的 serving 版本和公开权重可能不同。","DeepSeek harness 的工具策略需写入 system run。"],"fact_source_ids":["src-deepseek-v4","src-deepseek-updates"],"last_checked":"2026-08-27"},"sourceIds":["src-deepseek-v4","src-deepseek-updates"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"deepseek/v4-flash@2026-07-31","canonicalId":"deepseek/v4-flash@2026-07-31","familyId":"deepseek/v4","name":"DeepSeek V4 Flash","provider":"DeepSeek","mark":"D","release":"2026-07-31","releaseDate":"2026-07-31","access":"api","status":"active","availability":"api","aliases":["deepseek-v4-flash","deepseek-v4-flash-0731"],"tags":["Flash","效率","开放权重","代码 Agent"],"summary":"V4 Flash public beta；0731 是同架构/规模的 repost-trained snapshot。","modalities":["text"],"contextWindow":1000000,"paramsTotal":284000000000,"paramsActive":13000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"DeepSeek V4 Flash 高吞吐/低成本 beta variant。","capabilities":["推理","代码","长上下文","代码 Agent"],"best_for":["高吞吐代码 agent","Flash/Fast 横向比较","成本敏感部署"],"endpoint_ids":["deepseek-v4-flash","deepseek-v4-flash-0731"],"availability":["api","hosted","open-weights"],"reasoning_modes":["reasoning"],"architecture":"MoE","context_note":"目录记录 1M context；0731 是 repost-trained snapshot。","parameter_note":"284B total / 13B active。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["public beta 和 snapshot 更新会影响可复现性；保留 observed date 与 endpoint。"],"fact_source_ids":["src-deepseek-v4","src-deepseek-updates"],"last_checked":"2026-08-27"},"sourceIds":["src-deepseek-v4","src-deepseek-updates"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":4},{"id":"deepseek/v3.2@2025-12-01","canonicalId":"deepseek/v3.2@2025-12-01","familyId":"deepseek/v3","name":"DeepSeek V3.2","provider":"DeepSeek","mark":"D","release":"2025-12-01","releaseDate":"2025-12-01","access":"api","status":"previous","availability":"api","aliases":["deepseek-v3.2","deepseek-chat"],"tags":["上一代","开放权重","代码"],"summary":"","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"DeepSeek V3.2 上一代通用/代码 baseline。","capabilities":["推理","代码","工具调用"],"best_for":["V4 代际比较","开放权重/服务历史时间线"],"endpoint_ids":["deepseek-v3.2","deepseek-chat"],"availability":["api","hosted","open-weights"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["同一 API alias 可能路由到更新 snapshot，历史成绩需保留抓取日期。"],"fact_source_ids":["src-deepseek-updates"],"last_checked":"2026-08-27"},"sourceIds":["src-deepseek-updates"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"deepseek/r1@0528","canonicalId":"deepseek/r1@0528","familyId":"deepseek/r1","name":"DeepSeek-R1-0528","provider":"DeepSeek","mark":"D","release":"2025-05-28","releaseDate":"2025-05-28","access":"open_weights","status":"previous","availability":"open_weights","aliases":["deepseek-r1-0528"],"tags":["历史","推理","开放权重"],"summary":"","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"DeepSeek-R1-0528 历史 reasoning release，保留作开放权重推理基线。","capabilities":["深度推理","数学","代码"],"best_for":["数学/推理时间线","开放权重 reasoning 对照"],"endpoint_ids":["DeepSeek-R1-0528"],"availability":["open-weights"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["模型卡、量化和推理框架不同会改变速度/显存；benchmark 仍需保留 protocol。"],"fact_source_ids":["deepseek-r1-0528"],"last_checked":"2026-08-27"},"sourceIds":["deepseek-r1-0528"],"scores":{"gpqa-diamond":{"value":68.4,"raw_value":"68.4%","setting":"R1-0528 · reported","sourceId":"deepseek-r1-0528","sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API · R1-0528 update","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-28","published_at":"2025-05-28","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"R1-0528 · reported"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-deepseek-r1-0528-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"aime-2025":{"value":87.5,"raw_value":"87.5%","setting":"R1-0528 · reported","sourceId":"deepseek-r1-0528","sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API · R1-0528 update","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-28","published_at":"2025-05-28","benchmark_version":"aime-2025@v1","version":"2025","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"R1-0528 · reported"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-deepseek-r1-0528-aime-2025","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"mmlu-pro":{"value":81.2,"raw_value":"81.2%","setting":"R1-0528 · reported","sourceId":"deepseek-r1-0528","sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API · R1-0528 update","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-05-28","published_at":"2025-05-28","benchmark_version":"mmlu-pro@v1","version":"v1","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"R1-0528 · reported"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-deepseek-r1-0528-mmlu-pro","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":3,"catalogOnly":false,"systemRunCount":0},{"id":"moonshot/kimi-k3@2026-07-16","canonicalId":"moonshot/kimi-k3@2026-07-16","familyId":"moonshot/kimi-k3","name":"Kimi K3","provider":"Moonshot AI","mark":"K","release":"2026-07-16","releaseDate":"2026-07-16","access":"api","status":"active","availability":"api","aliases":["k3","kimi-k3"],"tags":["旗舰","开放权重","多模态","代码 Agent"],"summary":"Kimi K3 2.8T open 3T-class model；官方 benchmark 需保留 Kimi Code/Claude Code/Codex harness。","modalities":["text","image","video"],"contextWindow":1000000,"paramsTotal":2800000000000,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false,"default_effort":"max"},"profile":{"positioning":"Kimi K3 开放 3T-class 旗舰，强调长程 coding、知识工作和 reasoning。","capabilities":["推理","代码","原生视觉","长上下文","Agent"],"best_for":["长程代码 agent","多模态知识工作","大参数开放权重比较"],"endpoint_ids":["k3","kimi-k3"],"availability":["api","hosted","open-weights"],"reasoning_modes":["max"],"architecture":"Kimi Delta Attention + Attention Residuals","context_note":"官方技术博客记录 1M native context，并强调 native vision。","parameter_note":"2.8T total；active 参数未公开。","license":null,"caveats":["Kimi Code、Claude Code、Codex harness 会产生不同 system subject。","K3 与 256K endpoint 是同一 release 的 context/quota 变体。"],"fact_source_ids":["src-kimi-k3","src-kimi-code"],"last_checked":"2026-08-27"},"sourceIds":["src-kimi-k3","src-kimi-code"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":2},{"id":"moonshot/kimi-k3-256k@2026-07-16","canonicalId":"moonshot/kimi-k3-256k@2026-07-16","familyId":"moonshot/kimi-k3","name":"Kimi K3 256k","provider":"Moonshot AI","mark":"K","release":"2026-07-16","releaseDate":"2026-07-16","access":"api","status":"active","availability":"api","aliases":["k3-256k","kimi-k3-256k"],"tags":["旗舰","开放权重","效率"],"summary":"与 K3 同一 release 的 256k context/quota endpoint variant。","modalities":["text","image","video"],"contextWindow":256000,"paramsTotal":2800000000000,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false,"context_variant":true},"profile":{"positioning":"Kimi K3 的 256K context/quota endpoint variant。","capabilities":["推理","代码","原生视觉","Agent"],"best_for":["成本/上下文受限的 K3 使用","与 1M endpoint 做 context 对照"],"endpoint_ids":["k3-256k","kimi-k3-256k"],"availability":["api","hosted","open-weights"],"reasoning_modes":["max"],"architecture":"同 Kimi K3 release","context_note":"256K context；不是新的训练权重。","parameter_note":"继承 K3 的 2.8T total 目录记录；active 未公开。","license":null,"caveats":["不要把 256K 与 1M endpoint 的结果合并成一个无条件分数。"],"fact_source_ids":["src-kimi-k3","src-kimi-code"],"last_checked":"2026-08-27"},"sourceIds":["src-kimi-k3","src-kimi-code"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"moonshot/kimi-k2.7-code@current","canonicalId":"moonshot/kimi-k2.7-code@current","familyId":"moonshot/kimi-k2.7","name":"Kimi K2.7 Code","provider":"Moonshot AI","mark":"K","release":null,"releaseDate":null,"access":"api","status":"active","availability":"api","aliases":["kimi-for-coding","kimi-k2.7-code"],"tags":["代码","效率","专用"],"summary":"","modalities":["text"],"contextWindow":256000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"specialization":"coding"},"profile":{"positioning":"Kimi Code 面向软件工程的 serving tier。","capabilities":["代码","仓库理解","工具调用","Agent"],"best_for":["代码 agent","终端/IDE 工作流","高吞吐编程"],"endpoint_ids":["kimi-for-coding","kimi-k2.7-code"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":"K2-family coding serving tier；精确 K2.7 物理架构未单独披露","context_note":"Kimi Code 官方表格记录 256,000-token context。","parameter_note":"K2.7 endpoint 未单独披露参数；Kimi K2 底座的 1.04T total / 32B active 仅作 lineage 参考。","parameter_evidence":[{"kind":"base_lineage","label":"Kimi K2 base lineage","total_b":1040,"active_b":32,"source_id":"src-kimi-k2-hf","confidence":"lineage only","note":"K2.7 是后续 coding derivative；底座参数不得写入 K2.7 endpoint 的 canonical fields。"}],"license":null,"caveats":["这是 coding serving tier；SWE/Terminal 分数不能冒充裸 K2.7 模型分数。","官方文档说明 non-thinking 请求可能路由到 K2.6；run 必须记录 thinking mode 与 resolved model。"],"fact_source_ids":["src-kimi-code","src-kimi-k2-hf"],"last_checked":"2026-08-27"},"sourceIds":["src-kimi-code","src-kimi-k2-hf"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"moonshot/kimi-k2.7-code-highspeed@current","canonicalId":"moonshot/kimi-k2.7-code-highspeed@current","familyId":"moonshot/kimi-k2.7","name":"Kimi K2.7 Code Highspeed","provider":"Moonshot AI","mark":"K","release":null,"releaseDate":null,"access":"api","status":"active","availability":"api","aliases":["kimi-for-coding-highspeed"],"tags":["代码","高速","效率"],"summary":"","modalities":["text"],"contextWindow":256000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"specialization":"coding","speed_variant":true},"profile":{"positioning":"Kimi Code 的 high-speed serving variant，优先吞吐和延迟。","capabilities":["代码","仓库理解","工具调用","Agent"],"best_for":["交互式 coding","高并发代码任务","Flash/Fast 对比"],"endpoint_ids":["kimi-for-coding-highspeed"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":"K2.7 Code 的 serving-speed variant；未披露独立权重变化","context_note":"Kimi Code 官方表格记录 256,000-token context。","parameter_note":"参数量未单独披露；K2 底座数据仅作 lineage 参考。","parameter_evidence":[{"kind":"base_lineage","label":"Kimi K2 base lineage","total_b":1040,"active_b":32,"source_id":"src-kimi-k2-hf","confidence":"lineage only","note":"Highspeed 是 serving variant；不把底座参数冒充为独立实测值。"}],"license":null,"caveats":["Highspeed 是 serving variant，不应新增或复用独立权重分数。"],"fact_source_ids":["src-kimi-code","src-kimi-k2-hf"],"last_checked":"2026-08-27"},"sourceIds":["src-kimi-code","src-kimi-k2-hf"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"moonshot/kimi-k2.6@2026-04-20","canonicalId":"moonshot/kimi-k2.6@2026-04-20","familyId":"moonshot/kimi-k2","name":"Kimi K2.6","provider":"Moonshot AI","mark":"K","release":"2026-04-20","releaseDate":"2026-04-20","access":"api","status":"previous","availability":"api","aliases":["kimi-k2.6","kimi-k2-6"],"tags":["上一代","开放权重","Agent"],"summary":"","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Kimi K2.6 上一代开放权重/服务模型，覆盖代码、视觉和 agent。","capabilities":["代码","视觉","推理","Agent"],"best_for":["Kimi 代际比较","多模态代码 agent"],"endpoint_ids":["kimi-k2.6"],"availability":["api","hosted","open-weights"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","license":null,"caveats":["不同 Kimi Code 配置和开放权重 checkpoint 需分开记录。"],"fact_source_ids":["src-kimi-code"],"last_checked":"2026-08-27"},"sourceIds":["src-kimi-code"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"zai/glm-5.3@2026-08-14","canonicalId":"zai/glm-5.3@2026-08-14","familyId":"zai/glm-5","name":"GLM-5.3","provider":"Z.ai","mark":"Z","release":"2026-08-14","releaseDate":"2026-08-14","access":"api","status":"preview","availability":"api","aliases":["glm-5.3"],"tags":["旗舰","代码 Agent","推理"],"summary":"GLM-5.3 weights release was announced for a later date; keep preview/availability separate from reported scores.","modalities":["text"],"contextWindow":1000000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":true,"effort":"max"},"profile":{"positioning":"GLM-5.3 前沿预览，聚焦代码 agent 与复杂 reasoning。","capabilities":["推理","代码","代码 Agent","工具调用"],"best_for":["预览 frontier coding","GLM 代际比较"],"endpoint_ids":["glm-5.3"],"availability":["api","hosted-preview"],"reasoning_modes":["max"],"architecture":"与 GLM-5.2 相同的 base architecture；本次为 post-training update","context_note":"官方 GLM-5.3 blog 记录 1,000,000-token preview context。","parameter_note":"GLM-5.3 未单独发布权重计数；GLM-5.2 的约 753B total / 40B active 仅作 base-lineage 参考。","parameter_evidence":[{"kind":"base_lineage","label":"GLM-5.2 base lineage","total_b":753,"active_b":40,"approximate":true,"source_id":"src-glm53","confidence":"vendor-described lineage","note":"官方仅说明 GLM-5.3 沿用 GLM-5.2 底座并通过 post-training 提升；不写入 GLM-5.3 canonical 参数。"}],"license":null,"caveats":["预览服务与后续 weights release 可能不是同一 snapshot。","榜单 reported 值必须保留发布日期和 endpoint。"],"fact_source_ids":["src-glm53"],"last_checked":"2026-08-27"},"sourceIds":["src-glm53"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":3},{"id":"zai/glm-5.3-flash@2026-08-27","canonicalId":"zai/glm-5.3-flash@2026-08-27","familyId":"zai/glm-5","name":"GLM-5.3 Flash","provider":"Z.ai","mark":"Z","release":"2026-08-27","releaseDate":"2026-08-27","access":"open_weights","status":"active","availability":"open_weights","aliases":["glm-5.3-flash"],"tags":["Flash","效率","代码"],"summary":"","modalities":["text","image"],"contextWindow":1000000,"paramsTotal":320000000000,"paramsActive":18000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false,"speed_variant":true},"profile":{"positioning":"GLM-5.3 Flash 低延迟/低成本 serving variant。","capabilities":["推理","代码","工具调用"],"best_for":["Flash 横向比较","高吞吐代码和工具任务"],"endpoint_ids":["glm-5.3-flash"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":"Native multimodal MoE；hybrid sparse/linear attention + mHC","context_note":"官方规格记录 1,000,000-token context。","parameter_note":"官方 exact-model 规格披露 320B total / 18B active per token。","license":"MIT","caveats":["Flash 速度/成本 tier 与 GLM-5.3 主模型不应直接合并排名。"],"fact_source_ids":["src-glm53-flash","src-glm53-flash-hf"],"last_checked":"2026-08-27"},"sourceIds":["src-glm53-flash","src-glm53-flash-hf"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"zai/glm-5.2@2026-06-16","canonicalId":"zai/glm-5.2@2026-06-16","familyId":"zai/glm-5","name":"GLM-5.2","provider":"Z.ai","mark":"Z","release":"2026-06-16","releaseDate":"2026-06-16","access":"open_weights","status":"previous","availability":"open_weights","aliases":["glm-5.2","glm-5-2"],"tags":["上一代","开放权重","代码"],"summary":"","modalities":["text"],"contextWindow":1000000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GLM-5.2 上一代开放权重/托管 baseline。","capabilities":["推理","代码","长上下文","工具调用"],"best_for":["GLM 5.3 代际对照","开放权重长上下文比较"],"endpoint_ids":["glm-5.2"],"availability":["api","hosted","open-weights"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"目录记录 1M context；部署端限制需按 provider/权重版本区分。","parameter_note":"参数量未公开。","license":null,"caveats":["公开权重与 hosted endpoint 的 tokenizer/serving 配置可能不同。"],"fact_source_ids":["src-glm52"],"last_checked":"2026-08-27"},"sourceIds":["src-glm52"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"zai/glm-5.1@2026-04-07","canonicalId":"zai/glm-5.1@2026-04-07","familyId":"zai/glm-5","name":"GLM-5.1","provider":"Z.ai","mark":"Z","release":"2026-04-07","releaseDate":"2026-04-07","access":"api","status":"previous","availability":"api","aliases":["glm-5.1","glm-5-1"],"tags":["上一代","代码 Agent"],"summary":"","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GLM-5.1 代码/Agent 上一代 baseline。","capabilities":["推理","代码","代码 Agent","工具调用"],"best_for":["GLM 代际时间线","软件工程 agent 对照"],"endpoint_ids":["glm-5.1"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","license":null,"caveats":["Agent benchmark 必须记录工具和 scaffold。"],"fact_source_ids":["src-glm52"],"last_checked":"2026-08-27"},"sourceIds":["src-glm52"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"zai/glm-5@2026-02-12","canonicalId":"zai/glm-5@2026-02-12","familyId":"zai/glm-5","name":"GLM-5","provider":"Z.ai","mark":"Z","release":"2026-02-12","releaseDate":"2026-02-12","access":"open_weights","status":"previous","availability":"open_weights","aliases":["glm-5"],"tags":["历史","开放权重","代码"],"summary":"","modalities":["text"],"contextWindow":null,"paramsTotal":744000000000,"paramsActive":40000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GLM-5 早期开放权重 Agent/reasoning baseline。","capabilities":["推理","代码","Agent","工具调用"],"best_for":["开放权重规模比较","GLM 历史 benchmark"],"endpoint_ids":["glm-5"],"availability":["api","hosted","open-weights"],"reasoning_modes":["reasoning"],"architecture":"MoE","context_note":"当前目录未统一核验 context。","parameter_note":"744B total / 40B active。","license":null,"caveats":["参数口径仅适用于来源所述 checkpoint；量化不会改变基础参数量。"],"fact_source_ids":["src-glm52"],"last_checked":"2026-08-27"},"sourceIds":["src-glm52"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"minimax/m3@2026-05-31","canonicalId":"minimax/m3@2026-05-31","familyId":"minimax/m3","name":"MiniMax M3","provider":"MiniMax","mark":"M","release":"2026-05-31","releaseDate":"2026-05-31","access":"api","status":"active","availability":"api","aliases":["MiniMax-M3","m3"],"tags":["旗舰","多模态","Agent"],"summary":"","modalities":["text","image"],"contextWindow":1000000,"paramsTotal":428000000000,"paramsActive":23000000000,"paramsApproximate":true,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"MiniMax M3 coding/agentic flagship，采用 Sparse Attention 并支持多模态。","capabilities":["代码","Agent","推理","视觉","长视频理解"],"best_for":["长程代码 agent","多模态 agent","1M context 工作流"],"endpoint_ids":["MiniMax-M3","m3"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":"Native multimodal MoE + MiniMax Sparse Attention (MSA)","context_note":"官方模型页记录最多 1M、保证至少 512K context。","parameter_note":"官方 model card 披露约 428B total / 23B active；目录以 approximate 标记，不显示为精确计数。","license":"minimax-community","caveats":["长视频/agent 结果依赖工具、上下文截断策略和 serving 配置。"],"fact_source_ids":["src-minimax-m3","src-minimax-release","src-minimax-m3-hf"],"last_checked":"2026-08-27"},"sourceIds":["src-minimax-m3","src-minimax-release","src-minimax-m3-hf"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"minimax/m2.7@2026-03-18","canonicalId":"minimax/m2.7@2026-03-18","familyId":"minimax/m2","name":"MiniMax M2.7","provider":"MiniMax","mark":"M","release":"2026-03-18","releaseDate":"2026-03-18","access":"api","status":"previous","availability":"api","aliases":["MiniMax-M2.7","m2.7","minimax-m2-7"],"tags":["上一代","代码 Agent"],"summary":"","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"MiniMax M2.7 上一代代码/Agent serving release。","capabilities":["代码","Agent","推理","工具调用"],"best_for":["M3 代际比较","代码 agent 历史 baseline"],"endpoint_ids":["MiniMax-M2.7","m2.7"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"当前公开 release notes 未统一核验 context。","parameter_note":"参数量未公开。","license":null,"caveats":["M2.7 与 HighSpeed 是不同 serving 条件，需分行保存。"],"fact_source_ids":["src-minimax-release"],"last_checked":"2026-08-27"},"sourceIds":["src-minimax-release"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"minimax/m2.7-highspeed@2026-03-18","canonicalId":"minimax/m2.7-highspeed@2026-03-18","familyId":"minimax/m2","name":"MiniMax M2.7 Highspeed","provider":"MiniMax","mark":"M","release":"2026-03-18","releaseDate":"2026-03-18","access":"api","status":"previous","availability":"api","aliases":["MiniMax-M2.7-highspeed"],"tags":["上一代","高速","效率"],"summary":"","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"speed_variant":true},"profile":{"positioning":"MiniMax M2.7 的 high-speed serving variant。","capabilities":["代码","Agent","工具调用"],"best_for":["交互式 coding","高吞吐 agent","速度/成本对比"],"endpoint_ids":["MiniMax-M2.7-highspeed"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":"同 MiniMax M2.7 serving family","context_note":"HighSpeed endpoint 的 context/配额需按 release notes 复核。","parameter_note":"继承 M2.7 的公开参数口径（若有）；本目录未统一核验。","license":null,"caveats":["速度 variant 不是新训练 release；不要重复计入模型族规模。"],"fact_source_ids":["src-minimax-release"],"last_checked":"2026-08-27"},"sourceIds":["src-minimax-release"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"minimax/m2.5@2026-02","canonicalId":"minimax/m2.5@2026-02","familyId":"minimax/m2","name":"MiniMax M2.5","provider":"MiniMax","mark":"M","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["MiniMax-M2.5","m2.5"],"tags":["历史","代码 Agent"],"summary":"","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"MiniMax M2.5 历史生产 baseline。","capabilities":["代码","Agent","推理"],"best_for":["MiniMax 代际时间线","历史代码 agent 对照"],"endpoint_ids":["MiniMax-M2.5","m2.5"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","license":null,"caveats":["release_label 为 2026-02；具体发布日期和 alias 需要后续从 release notes 补齐。"],"fact_source_ids":["src-minimax-release"],"last_checked":"2026-08-27"},"sourceIds":["src-minimax-release"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"xiaomi/mimo-v2.5-pro@2026-04-23","canonicalId":"xiaomi/mimo-v2.5-pro@2026-04-23","familyId":"xiaomi/mimo-v2.5","name":"MiMo-V2.5-Pro","provider":"Xiaomi MiMo","mark":"X","release":"2026-04-23","releaseDate":"2026-04-23","access":"api","status":"active","availability":"api","aliases":["MiMo-V2.5-Pro","mimo-v2.5-pro"],"tags":["旗舰","MoE","推理","Agent"],"summary":"","modalities":["text","image"],"contextWindow":1000000,"paramsTotal":1000000000000,"paramsActive":42000000000,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"MiMo-V2.5-Pro 万亿级 MoE 旗舰，面向推理和复杂 agent 工作。","capabilities":["深度推理","代码","Agent","视觉","长上下文"],"best_for":["复杂 agent workflow","长文档/代码任务","万亿参数规模对比"],"endpoint_ids":["mimo-v2.5-pro"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":"Mixture-of-Experts, 1T total / 42B active","context_note":"官方 MiMo model docs 记录 1M context。","parameter_note":"1T total / 42B active。","license":"MIT","caveats":["Pro 与 UltraSpeed 是同 release 的不同 serving/deployment 条件。","Agent 成绩需记录是否使用 OpenClaw/工具集成。"],"fact_source_ids":["src-mimo-models","src-mimo-updates"],"last_checked":"2026-08-27"},"sourceIds":["src-mimo-models","src-mimo-updates"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"xiaomi/mimo-v2.5@2026-04-23","canonicalId":"xiaomi/mimo-v2.5@2026-04-23","familyId":"xiaomi/mimo-v2.5","name":"MiMo-V2.5","provider":"Xiaomi MiMo","mark":"X","release":"2026-04-23","releaseDate":"2026-04-23","access":"api","status":"active","availability":"api","aliases":["MiMo-V2.5","mimo-v2.5","mimo-v2-5"],"tags":["多模态","Agent","效率"],"summary":"","modalities":["text","image","video","audio"],"contextWindow":1000000,"paramsTotal":309000000000,"paramsActive":15000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false,"omni":true},"profile":{"positioning":"MiMo-V2.5 omni-modal production model，覆盖文本、图像、视频和音频。","capabilities":["全模态理解","语音","视频","Agent","长上下文"],"best_for":["多模态/语音应用","视频理解","跨模态 agent"],"endpoint_ids":["mimo-v2.5"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":"Mixture-of-Experts","context_note":"官方 MiMo model docs 记录 1M context。","parameter_note":"官方 checkpoint model card 披露 309B semantic total / 15B active；tensor count 约 311B，二者口径分开。","license":"MIT","caveats":["不同输入模态的 API 处理、采样和费用不同；benchmark 必须标注 modality。"],"fact_source_ids":["src-mimo-models","src-mimo-updates","src-mimo-v25-hf"],"last_checked":"2026-08-27"},"sourceIds":["src-mimo-models","src-mimo-updates","src-mimo-v25-hf"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"xiaomi/mimo-v2.5-pro-ultraspeed@current","canonicalId":"xiaomi/mimo-v2.5-pro-ultraspeed@current","familyId":"xiaomi/mimo-v2.5","name":"MiMo-V2.5-Pro UltraSpeed","provider":"Xiaomi MiMo","mark":"X","release":null,"releaseDate":null,"access":"api","status":"active","availability":"api","aliases":["MiMo-V2.5-Pro UltraSpeed"],"tags":["高速","效率","多模态"],"summary":"","modalities":["text","image"],"contextWindow":1000000,"paramsTotal":1000000000000,"paramsActive":42000000000,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"speed_variant":true},"profile":{"positioning":"MiMo-V2.5-Pro 的 UltraSpeed 部署 variant，优化吞吐而非新增权重。","capabilities":["推理","代码","Agent","视觉","高速推理"],"best_for":["低延迟 agent","高吞吐 API","部署/速度比较"],"endpoint_ids":["mimo-v2.5-pro-ultraspeed"],"availability":["api","early-access"],"reasoning_modes":["reasoning"],"architecture":"同 MiMo-V2.5-Pro；FP4 + DFlash serving optimization","context_note":"目录记录 1M context；UltraSpeed 的实际配额需按当前服务核验。","parameter_note":"继承 Pro 的 1T total / 42B active 目录口径。","license":null,"caveats":["FP4 lossless quantization、DFlash 和 TileRT 是部署优化，不应当作新 model release。"],"fact_source_ids":["src-mimo-models"],"last_checked":"2026-08-27"},"sourceIds":["src-mimo-models"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"google/gemini-3.1-pro@2026-02-19","canonicalId":"google/gemini-3.1-pro@2026-02-19","familyId":"google/gemini-3.1","name":"Gemini 3.1 Pro","provider":"Google","mark":"G","release":"2026-02-19","releaseDate":"2026-02-19","access":"api","status":"preview","availability":"api","aliases":["gemini-3.1-pro","gemini-3-1-pro-preview"],"tags":["多模态","推理","Agent"],"summary":"Gemini 3.1 Pro preview endpoint；1,048,576-token input、65,536-token max output，多模态原生能力。","modalities":["text","image","video","audio"],"contextWindow":1048576,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":true},"profile":{"positioning":"Gemini 3.1 Pro 多模态 reasoning baseline，覆盖长上下文和 agent 场景。","capabilities":["推理","文本","图像","视频","音频","工具调用"],"best_for":["多模态知识工作","Google 生态工具任务","跨模态 reasoning"],"endpoint_ids":["gemini-3.1-pro","gemini-3-1-pro-preview"],"availability":["api","hosted"],"reasoning_modes":["thinking"],"architecture":"Native multimodal model based on Gemini 3 Pro（官方高层描述）","context_note":"官方 preview endpoint 规格记录 1,048,576 input tokens。","parameter_note":"参数量未公开。","max_output_tokens":65536,"license":null,"caveats":["Preview alias 与稳定 endpoint 可能路由不同；保留 source snapshot 和日期。"],"fact_source_ids":["google-gemini31","src-google-gemini31-spec"],"last_checked":"2026-08-27"},"sourceIds":["google-gemini31","src-google-gemini31-spec"],"scores":{"gpqa-diamond":{"value":94.3,"raw_value":"94.3%","setting":"no tools · thinking high","sourceId":"google-gemini31","sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","sourceLabel":"Google DeepMind · Gemini 3.1 Pro model card","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-02-19","published_at":"2026-02-19","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools · thinking high"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-3-1-pro-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"mmmu":{"value":80.5,"raw_value":"80.5%","setting":"MMMU-Pro · no tools","sourceId":"google-gemini31","sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","sourceLabel":"Google DeepMind · Gemini 3.1 Pro model card","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-02-19","published_at":"2026-02-19","benchmark_version":"mmmu@v1","version":"original","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU-Pro · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-3-1-pro-mmmu","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"hle":{"value":44.4,"raw_value":"44.4%","setting":"full set · no tools","sourceId":"google-gemini31","sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","sourceLabel":"Google DeepMind · Gemini 3.1 Pro model card","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2026-02-19","published_at":"2026-02-19","benchmark_version":"hle@v1","version":"full","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"full set · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-3-1-pro-hle","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":3,"catalogOnly":false,"systemRunCount":2},{"id":"google/gemini-2.5-pro@2025-06-05","canonicalId":"google/gemini-2.5-pro@2025-06-05","familyId":"google/gemini-2.5","name":"Gemini 2.5 Pro","provider":"Google","mark":"G","release":"2025-06-05","releaseDate":"2025-06-05","access":"api","status":"previous","availability":"api","aliases":["gemini-2.5-pro"],"tags":["上一代","多模态","长上下文"],"summary":"","modalities":["text","image","video","audio"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Gemini 2.5 Pro 上一代多模态 reasoning baseline。","capabilities":["推理","文本","图像","视频","音频","长上下文"],"best_for":["Gemini 3.1 代际比较","多模态历史 benchmark"],"endpoint_ids":["gemini-2.5-pro"],"availability":["api","hosted"],"reasoning_modes":["thinking"],"architecture":null,"context_note":"当前目录未统一核验 context；不同 Google API surface 可能有不同上限。","parameter_note":"参数量未公开。","license":null,"caveats":["Gemini benchmark 常区分 thinking budget、tools 和 modality；不可只按显示名合并。"],"fact_source_ids":["google-gemini25"],"last_checked":"2026-08-27"},"sourceIds":["google-gemini25"],"scores":{"gpqa-diamond":{"value":86.4,"raw_value":"86.4%","setting":"single attempt · no tools","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-06-05","published_at":"2025-06-05","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"single attempt · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-2-5-pro-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"aime-2025":{"value":88.0,"raw_value":"88.0%","setting":"single attempt · no tools","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-06-05","published_at":"2025-06-05","benchmark_version":"aime-2025@v1","version":"2025","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"single attempt · no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-2-5-pro-aime-2025","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"livecodebench":{"value":70.4,"raw_value":"70.4%","setting":"V5 · rolling window","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-03-25","published_at":"2025-03-25","benchmark_version":"livecodebench@v5","version":"V5","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"V5 · rolling window"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-2-5-pro-livecodebench","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"mmmu":{"value":81.7,"raw_value":"81.7%","setting":"MMMU · single attempt","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-06-05","published_at":"2025-06-05","benchmark_version":"mmmu@v1","version":"original","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU · single attempt"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-2-5-pro-mmmu","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"hle":{"value":18.8,"raw_value":"18.8%","setting":"no tools","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-03-25","published_at":"2025-03-25","benchmark_version":"hle@v1","version":"full","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-gemini-2-5-pro-hle","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":5,"catalogOnly":false,"systemRunCount":1},{"id":"google/gemini-2.5-flash@current","canonicalId":"google/gemini-2.5-flash@current","familyId":"google/gemini-2.5","name":"Gemini 2.5 Flash","provider":"Google","mark":"G","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["gemini-2.5-flash"],"tags":["Flash","上一代","效率","多模态"],"summary":"","modalities":["text","image","video","audio"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"speed_variant":true},"profile":{"positioning":"Gemini 2.5 Flash 上一代快速/成本敏感多模态 tier。","capabilities":["快速推理","文本","图像","视频","音频","工具调用"],"best_for":["高吞吐多模态应用","Flash/Fast 横向比较","实时交互"],"endpoint_ids":["gemini-2.5-flash"],"availability":["api","hosted"],"reasoning_modes":["thinking"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","license":null,"caveats":["Flash 的 thinking budget 和 latency setting 需要写进 protocol。"],"fact_source_ids":["google-gemini25"],"last_checked":"2026-08-27"},"sourceIds":["google-gemini25"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"xai/grok-4-fast@2025-09-19","canonicalId":"xai/grok-4-fast@2025-09-19","familyId":"xai/grok-4","name":"Grok 4 Fast","provider":"xAI","mark":"X","release":"2025-09-19","releaseDate":"2025-09-19","access":"api","status":"previous","availability":"api","aliases":["grok-4-fast"],"tags":["上一代","实时","推理","效率"],"summary":"","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"speed_variant":true},"profile":{"positioning":"Grok 4 Fast 高吞吐、实时导向的上一代 serving tier。","capabilities":["推理","代码","实时信息","视觉输入"],"best_for":["实时问答","高吞吐 API","Grok 代际速度比较"],"endpoint_ids":["grok-4-fast"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","license":null,"caveats":["实时检索和工具开关会改变结果；需要区分裸模型与带 search 的 system。"],"fact_source_ids":["xai-grok4fast"],"last_checked":"2026-08-27"},"sourceIds":["xai-grok4fast"],"scores":{"gpqa-diamond":{"value":85.7,"raw_value":"85.7%","setting":"no tools","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-09-19","published_at":"2025-09-19","benchmark_version":"gpqa-diamond@v1","version":"Diamond","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-grok-4-fast-gpqa-diamond","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"aime-2025":{"value":92.0,"raw_value":"92.0%","setting":"no tools","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-09-19","published_at":"2025-09-19","benchmark_version":"aime-2025@v1","version":"2025","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-grok-4-fast-aime-2025","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"livecodebench":{"value":80.0,"raw_value":"80.0%","setting":"Jan–May rolling window","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-09-19","published_at":"2025-09-19","benchmark_version":"livecodebench@v5","version":"V5","metric":"pass@1","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"Jan–May rolling window"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-grok-4-fast-livecodebench","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},"hle":{"value":20.0,"raw_value":"20.0%","setting":"no tools","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast","verified":"reported","status":"reported","evidence_level":"B","comparability":"conditional","observed_at":"2025-09-19","published_at":"2025-09-19","benchmark_version":"hle@v1","version":"full","metric":"accuracy","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"harnessId":"model-only","harness":"Model-only / single turn","subjectType":"model","observationId":"legacy-grok-4-fast-hle","notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}},"scoreCount":4,"catalogOnly":false,"systemRunCount":0},{"id":"xai/grok-4@current","canonicalId":"xai/grok-4@current","familyId":"xai/grok-4","name":"Grok 4","provider":"xAI","mark":"X","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["grok-4"],"tags":["上一代","推理"],"summary":"","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Grok 4 通用 reasoning/代码模型，保留作为 xAI 历史 frontier 对照。","capabilities":["推理","代码","视觉输入","实时信息","工具调用"],"best_for":["Grok 代际比较","实时信息与工具任务"],"endpoint_ids":["grok-4"],"availability":["api","hosted"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"当前目录未统一核验 context。","parameter_note":"参数量未公开。","license":null,"caveats":["联网、系统提示和实时数据 snapshot 必须随成绩记录。"],"fact_source_ids":["xai-grok4fast"],"last_checked":"2026-08-27"},"sourceIds":["xai-grok4fast"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.1@2025-11-13","canonicalId":"openai/gpt-5.1@2025-11-13","familyId":"openai/gpt-5.1","name":"GPT-5.1","provider":"OpenAI","mark":"O","release":"2025-11-13","releaseDate":"2025-11-13","access":"api","status":"previous","availability":"api","aliases":["gpt-5.1","gpt-5.1-2025-11-13"],"tags":["上一代","推理","代码"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GPT-5.1：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","多模态"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["gpt-5.1","gpt-5.1-2025-11-13"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-openai-models","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.1-codex@2025-11-12","canonicalId":"openai/gpt-5.1-codex@2025-11-12","familyId":"openai/gpt-5.1-codex","name":"GPT-5.1 Codex","provider":"OpenAI","mark":"O","release":"2025-11-12","releaseDate":"2025-11-12","access":"api","status":"previous","availability":"api","aliases":["gpt-5.1-codex"],"tags":["代码","Agent"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GPT-5.1 Codex：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","工具调用"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["gpt-5.1-codex"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-openai-models","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.1-codex-max@2025-11-19","canonicalId":"openai/gpt-5.1-codex-max@2025-11-19","familyId":"openai/gpt-5.1-codex","name":"GPT-5.1 Codex Max","provider":"OpenAI","mark":"O","release":"2025-11-19","releaseDate":"2025-11-19","access":"api","status":"previous","availability":"api","aliases":["gpt-5.1-codex-max"],"tags":["代码","Agent","高计算"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GPT-5.1 Codex Max：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","工具调用"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["gpt-5.1-codex-max"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-openai-models","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.1-codex-mini@2025-11-17","canonicalId":"openai/gpt-5.1-codex-mini@2025-11-17","familyId":"openai/gpt-5.1-codex","name":"GPT-5.1 Codex Mini","provider":"OpenAI","mark":"O","release":"2025-11-17","releaseDate":"2025-11-17","access":"api","status":"previous","availability":"api","aliases":["gpt-5.1-codex-mini"],"tags":["代码","Agent","效率"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GPT-5.1 Codex Mini：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","工具调用"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["gpt-5.1-codex-mini"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-openai-models","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.2@2025-12-11","canonicalId":"openai/gpt-5.2@2025-12-11","familyId":"openai/gpt-5.2","name":"GPT-5.2","provider":"OpenAI","mark":"O","release":"2025-12-11","releaseDate":"2025-12-11","access":"api","status":"previous","availability":"api","aliases":["gpt-5.2","gpt-5.2-2025-12-11"],"tags":["推理","代码","工具调用"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GPT-5.2：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","多模态"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["gpt-5.2","gpt-5.2-2025-12-11"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-openai-models","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.2-pro@2025-12-11","canonicalId":"openai/gpt-5.2-pro@2025-12-11","familyId":"openai/gpt-5.2","name":"GPT-5.2 Pro","provider":"OpenAI","mark":"O","release":"2025-12-11","releaseDate":"2025-12-11","access":"api","status":"previous","availability":"api","aliases":["gpt-5.2-pro","gpt-5.2-pro-2025-12-11"],"tags":["推理","高计算"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GPT-5.2 Pro：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","多模态"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["gpt-5.2-pro","gpt-5.2-pro-2025-12-11"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-openai-models","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.2-codex@2025-12-18","canonicalId":"openai/gpt-5.2-codex@2025-12-18","familyId":"openai/gpt-5.2-codex","name":"GPT-5.2 Codex","provider":"OpenAI","mark":"O","release":"2025-12-18","releaseDate":"2025-12-18","access":"api","status":"previous","availability":"api","aliases":["gpt-5.2-codex"],"tags":["代码","Agent"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GPT-5.2 Codex：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","工具调用"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["gpt-5.2-codex"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-openai-models","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.3-codex@2026-02-05","canonicalId":"openai/gpt-5.3-codex@2026-02-05","familyId":"openai/gpt-5.3-codex","name":"GPT-5.3 Codex","provider":"OpenAI","mark":"O","release":"2026-02-05","releaseDate":"2026-02-05","access":"api","status":"active","availability":"api","aliases":["gpt-5.3-codex"],"tags":["代码","Agent","推理"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":400000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GPT-5.3 Codex：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","工具调用"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["gpt-5.3-codex"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"OpenAI endpoint 规格记录 400,000-token context。","parameter_note":"参数量未公开。","knowledge_cutoff":"2025-08-31","max_output_tokens":128000,"license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-openai-models","src-openai-gpt53-codex-spec","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-openai-models","src-openai-gpt53-codex-spec","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"anthropic/claude-opus-4.5@2025-11-01","canonicalId":"anthropic/claude-opus-4.5@2025-11-01","familyId":"anthropic/claude-opus","name":"Claude Opus 4.5","provider":"Anthropic","mark":"A","release":"2025-11-01","releaseDate":"2025-11-01","access":"api","status":"previous","availability":"api","aliases":["claude-opus-4-5-20251101","claude-opus-4.5"],"tags":["上一代","推理","代码","Agent"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Claude Opus 4.5：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","工具调用","多模态"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["claude-opus-4-5-20251101","claude-opus-4.5"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-anthropic-system-cards","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-anthropic-system-cards","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"anthropic/claude-opus-4.6@2026-02-05","canonicalId":"anthropic/claude-opus-4.6@2026-02-05","familyId":"anthropic/claude-opus","name":"Claude Opus 4.6","provider":"Anthropic","mark":"A","release":"2026-02-05","releaseDate":"2026-02-05","access":"api","status":"previous","availability":"api","aliases":["claude-opus-4-6","claude-opus-4.6"],"tags":["推理","代码","Agent"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Claude Opus 4.6：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","工具调用","多模态"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["claude-opus-4-6","claude-opus-4.6"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-anthropic-system-cards","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-anthropic-system-cards","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"anthropic/claude-sonnet-4.5@2025-09-29","canonicalId":"anthropic/claude-sonnet-4.5@2025-09-29","familyId":"anthropic/claude-sonnet","name":"Claude Sonnet 4.5","provider":"Anthropic","mark":"A","release":"2025-09-29","releaseDate":"2025-09-29","access":"api","status":"previous","availability":"api","aliases":["claude-sonnet-4-5-20250929","claude-sonnet-4.5"],"tags":["上一代","推理","代码"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Claude Sonnet 4.5：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","多模态"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["claude-sonnet-4-5-20250929","claude-sonnet-4.5"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-anthropic-system-cards","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-anthropic-system-cards","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"anthropic/claude-haiku-4.5@2025-10-01","canonicalId":"anthropic/claude-haiku-4.5@2025-10-01","familyId":"anthropic/claude-haiku","name":"Claude Haiku 4.5","provider":"Anthropic","mark":"A","release":"2025-10-01","releaseDate":"2025-10-01","access":"api","status":"previous","availability":"api","aliases":["claude-haiku-4-5-20251001","claude-haiku-4.5"],"tags":["效率","代码"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Claude Haiku 4.5：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","多模态"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["claude-haiku-4-5-20251001","claude-haiku-4.5"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-anthropic-system-cards","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-anthropic-system-cards","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.8-2.4t-a95b@2026-08-12","canonicalId":"qwen/qwen3.8-2.4t-a95b@2026-08-12","familyId":"qwen/qwen3.8","name":"Qwen3.8 2.4T A95B","provider":"Qwen","mark":"Q","release":"2026-08-12","releaseDate":"2026-08-12","access":"open_weights","status":"active","availability":"open_weights","aliases":["Qwen/Qwen3.8-2.4T-A95B","qwen3.8-2.4t-a95b"],"tags":["旗舰","开放权重","推理"],"summary":"Qwen3.8 open-weight 2.4T MoE checkpoint；95B active、262,144 native context，使用自定义 qwen3.8-max 许可。","modalities":["text"],"contextWindow":262144,"paramsTotal":2400000000000,"paramsActive":95000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.8 的公开权重 2.4T MoE checkpoint；95B parameters active per token。","capabilities":["推理"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["Qwen/Qwen3.8-2.4T-A95B","qwen3.8-2.4t-a95b"],"availability":["open-weights"],"reasoning_modes":["reasoning"],"architecture":"Mixture-of-Experts","context_note":"官方 model card 记录 262,144 native context；扩展部署配置需单独标注。","parameter_note":"官方 checkpoint 名称与 model card 直接披露 2.4T total / 95B active。","license":"qwen3.8-max custom license","caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-qwen38-repo","src-qwen38-24t-hf"],"last_checked":"2026-08-27"},"sourceIds":["src-qwen38-repo","src-qwen38-24t-hf"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.5-397b-a17b@2026-02-13","canonicalId":"qwen/qwen3.5-397b-a17b@2026-02-13","familyId":"qwen/qwen3.5","name":"Qwen3.5 397B A17B","provider":"Qwen","mark":"Q","release":"2026-02-13","releaseDate":"2026-02-13","access":"open_weights","status":"previous","availability":"open_weights","aliases":["qwen3.5-397b-a17b"],"tags":["上一代","MoE","推理"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.5 系列的 open-weight 397B/A17B 标识条目；公开报告只归到这个明确 checkpoint 名称。","capabilities":["文本生成","推理"],"best_for":["Qwen3.5 open-weight 代际对照","同代 MoE 规模档位比较"],"endpoint_ids":["qwen3.5-397b-a17b"],"availability":["open-weights"],"reasoning_modes":["reasoning"],"architecture":"MoE（结构细节待逐型号来源核验）","context_note":"该 checkpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"397B/A17B 来自型号名；在参数 scope 与 locator 补齐前不提升为机器参数事实。","license":null,"caveats":["型号名中的 total/active 标识仍需 source-backed parameter scope。","量化、搜索、effort 与 harness 结果必须保留各自 protocol。"],"fact_source_ids":["src-qwen36","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen36","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.5-122b-a10b@2026-02-13","canonicalId":"qwen/qwen3.5-122b-a10b@2026-02-13","familyId":"qwen/qwen3.5","name":"Qwen3.5 122B A10B","provider":"Qwen","mark":"Q","release":"2026-02-13","releaseDate":"2026-02-13","access":"open_weights","status":"previous","availability":"open_weights","aliases":["qwen3.5-122b-a10b"],"tags":["上一代","MoE","推理"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.5 系列的 open-weight 122B/A10B 标识条目，处在 397B/A17B 与 27B 档之间。","capabilities":["文本生成","推理"],"best_for":["Qwen3.5 open-weight 档位比较","MoE 部署候选的能力追踪"],"endpoint_ids":["qwen3.5-122b-a10b"],"availability":["open-weights"],"reasoning_modes":["reasoning"],"architecture":"MoE（结构细节待逐型号来源核验）","context_note":"该 checkpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"122B/A10B 来自型号名；在参数 scope 与 locator 补齐前不提升为机器参数事实。","license":null,"caveats":["型号名中的 total/active 标识仍需 source-backed parameter scope。","不能把不同量化或 serving 配置的吞吐归因于基础权重。"],"fact_source_ids":["src-qwen36","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen36","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.5-27b@2026-02-24","canonicalId":"qwen/qwen3.5-27b@2026-02-24","familyId":"qwen/qwen3.5","name":"Qwen3.5 27B","provider":"Qwen","mark":"Q","release":"2026-02-24","releaseDate":"2026-02-24","access":"open_weights","status":"previous","availability":"open_weights","aliases":["qwen3.5-27b"],"tags":["上一代","小模型","推理"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.5 系列的 27B open-weight 条目，作为较小部署档位保留。","capabilities":["文本生成","推理"],"best_for":["本地部署候选比较","Qwen 27B 代际对照"],"endpoint_ids":["qwen3.5-27b"],"availability":["open-weights"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该 checkpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"27B 来自型号名；parameter scope 与精确计数尚未在目录中逐项核验。","license":null,"caveats":["较小档位不等于已验证低成本；VRAM、量化和吞吐需 deployment evidence。","不同 Qwen 27B 代际不能按规模名自动合并。"],"fact_source_ids":["src-qwen36","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen36","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.5-plus@2026-02-16","canonicalId":"qwen/qwen3.5-plus@2026-02-16","familyId":"qwen/qwen3.5","name":"Qwen3.5 Plus","provider":"Qwen","mark":"Q","release":"2026-02-16","releaseDate":"2026-02-16","access":"api","status":"previous","availability":"api","aliases":["qwen3.5-plus"],"tags":["上一代","推理"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.5 Plus 上一代托管多模态 reasoning endpoint。","capabilities":["推理","文本生成","视觉输入"],"best_for":["Qwen Plus 托管端点代际对照","图文 reasoning 报告追踪"],"endpoint_ids":["qwen3.5-plus"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该托管 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"托管 endpoint 参数量未公开。","license":null,"caveats":["Plus 服务名不能自动映射到名称相近的 open-weight checkpoint。","视觉与纯文本结果必须按输入模态分列。"],"fact_source_ids":["src-qwen36","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen36","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.5-flash@2026-02-25","canonicalId":"qwen/qwen3.5-flash@2026-02-25","familyId":"qwen/qwen3.5","name":"Qwen3.5 Flash","provider":"Qwen","mark":"Q","release":"2026-02-25","releaseDate":"2026-02-25","access":"api","status":"previous","availability":"api","aliases":["qwen3.5-flash"],"tags":["Flash","效率"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.5 Flash 上一代托管效率 endpoint；Flash 是产品层级标签，不是实测速度结论。","capabilities":["文本生成","推理"],"best_for":["Qwen Flash 代际对照","质量/成本/延迟 Pareto 候选"],"endpoint_ids":["qwen3.5-flash"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该托管 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"托管 endpoint 参数量未公开。","license":null,"caveats":["Flash 标签不能替代 TTFT、tokens/s、价格和 batch 条件。","不能把该服务 endpoint 当作已公开权重。"],"fact_source_ids":["src-qwen36","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen36","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.6-plus@2026-03-31","canonicalId":"qwen/qwen3.6-plus@2026-03-31","familyId":"qwen/qwen3.6","name":"Qwen3.6 Plus","provider":"Qwen","mark":"Q","release":"2026-03-31","releaseDate":"2026-03-31","access":"api","status":"previous","availability":"api","aliases":["qwen3.6-plus"],"tags":["上一代","推理"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.6 Plus 上一代托管多模态 reasoning endpoint，位于 3.5 Plus 与 3.7 Plus 之间。","capabilities":["推理","文本生成","视觉输入"],"best_for":["Qwen Plus 连续代际对照","托管图文 reasoning 报告追踪"],"endpoint_ids":["qwen3.6-plus"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该托管 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"托管 endpoint 参数量未公开。","license":null,"caveats":["Plus endpoint 不自动继承同代 open-weight checkpoint 的参数事实。","视觉与纯文本结果必须按输入模态分列。"],"fact_source_ids":["src-qwen36","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen36","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.6-27b@2026-04-22","canonicalId":"qwen/qwen3.6-27b@2026-04-22","familyId":"qwen/qwen3.6","name":"Qwen3.6 27B","provider":"Qwen","mark":"Q","release":"2026-04-22","releaseDate":"2026-04-22","access":"open_weights","status":"previous","availability":"open_weights","aliases":["qwen3.6-27b"],"tags":["小模型","开放权重"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.6 系列的 27B open-weight 条目，用于较小部署档位和 27B 代际比较。","capabilities":["文本生成","推理"],"best_for":["本地部署候选比较","Qwen 27B 代际对照"],"endpoint_ids":["qwen3.6-27b"],"availability":["open-weights"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该 checkpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"27B 来自型号名；parameter scope 与精确计数尚未在目录中逐项核验。","license":null,"caveats":["较小档位不等于已验证低成本；VRAM、量化和吞吐需 deployment evidence。","不同 Qwen 27B 代际不能按规模名自动合并。"],"fact_source_ids":["src-qwen36","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen36","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.6-35b-a3b@2026-04-14","canonicalId":"qwen/qwen3.6-35b-a3b@2026-04-14","familyId":"qwen/qwen3.6","name":"Qwen3.6 35B A3B","provider":"Qwen","mark":"Q","release":"2026-04-14","releaseDate":"2026-04-14","access":"open_weights","status":"previous","availability":"open_weights","aliases":["qwen3.6-35b-a3b"],"tags":["MoE","效率"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.6 系列的 35B/A3B open-weight MoE 标识条目，作为低 active-parameter 候选。","capabilities":["文本生成","推理"],"best_for":["MoE 部署效率候选比较","Qwen3.6 open-weight 档位对照"],"endpoint_ids":["qwen3.6-35b-a3b"],"availability":["open-weights"],"reasoning_modes":["reasoning"],"architecture":"MoE（结构细节待逐型号来源核验）","context_note":"该 checkpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"35B/A3B 来自型号名；在参数 scope 与 locator 补齐前不提升为机器参数事实。","license":null,"caveats":["active 参数标签不能替代实测显存、吞吐或成本。","不同量化与推理框架必须作为 deployment 条件记录。"],"fact_source_ids":["src-qwen36","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen36","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.6-flash@2026-04-27","canonicalId":"qwen/qwen3.6-flash@2026-04-27","familyId":"qwen/qwen3.6","name":"Qwen3.6 Flash","provider":"Qwen","mark":"Q","release":"2026-04-27","releaseDate":"2026-04-27","access":"api","status":"previous","availability":"api","aliases":["qwen3.6-flash"],"tags":["Flash","效率"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.6 Flash 上一代托管效率 endpoint，用于 Qwen Flash 时间线。","capabilities":["文本生成","推理"],"best_for":["Qwen Flash 代际对照","质量/成本/延迟 Pareto 候选"],"endpoint_ids":["qwen3.6-flash"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该托管 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"托管 endpoint 参数量未公开。","license":null,"caveats":["Flash 标签不能替代 TTFT、tokens/s、价格和 batch 条件。","该 endpoint 与 open-weight Flash-Next 不应自动合并。"],"fact_source_ids":["src-qwen36","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen36","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.7-max@2026-05-19","canonicalId":"qwen/qwen3.7-max@2026-05-19","familyId":"qwen/qwen3.7","name":"Qwen3.7 Max","provider":"Qwen","mark":"Q","release":"2026-05-19","releaseDate":"2026-05-19","access":"api","status":"previous","availability":"api","aliases":["qwen3.7-max"],"tags":["上一代","旗舰","推理"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.7 Max 上一代托管旗舰 endpoint，保留作 3.8 Max 的服务代际基线。","capabilities":["推理","文本生成","视觉输入"],"best_for":["Qwen Max 代际对照","托管多模态旗舰报告追踪"],"endpoint_ids":["qwen3.7-max"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该托管 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"托管 endpoint 参数量未公开。","license":null,"caveats":["Max 服务名不能自动映射到名称相近的 open-weight checkpoint。","服务路由或 snapshot 变化会影响跨日期比较。"],"fact_source_ids":["src-qwen36","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen36","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.7-flash@2026-07-27","canonicalId":"qwen/qwen3.7-flash@2026-07-27","familyId":"qwen/qwen3.7","name":"Qwen3.7 Flash","provider":"Qwen","mark":"Q","release":"2026-07-27","releaseDate":"2026-07-27","access":"api","status":"active","availability":"api","aliases":["qwen3.7-flash"],"tags":["Flash","效率"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.7 Flash 当前托管效率 endpoint，用于 Flash/Fast 质量与服务指标对照。","capabilities":["文本生成","推理"],"best_for":["Qwen Flash 代际对照","质量/成本/延迟 Pareto 候选"],"endpoint_ids":["qwen3.7-flash"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该托管 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"托管 endpoint 参数量未公开。","license":null,"caveats":["Flash 标签不能替代 TTFT、tokens/s、价格和 batch 条件。","不能把不同日期或路由的 hosted snapshot 静默合并。"],"fact_source_ids":["src-qwen36","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen36","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"deepseek/v3.1@2025-08-21","canonicalId":"deepseek/v3.1@2025-08-21","familyId":"deepseek/v3","name":"DeepSeek V3.1","provider":"DeepSeek","mark":"D","release":"2025-08-21","releaseDate":"2025-08-21","access":"api","status":"previous","availability":"api","aliases":["deepseek-v3.1"],"tags":["上一代","推理","代码"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"DeepSeek V3.1：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["deepseek-v3.1"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-deepseek-updates","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-deepseek-updates","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"deepseek/v3.1-terminus@2025-09-22","canonicalId":"deepseek/v3.1-terminus@2025-09-22","familyId":"deepseek/v3","name":"DeepSeek V3.1 Terminus","provider":"DeepSeek","mark":"D","release":"2025-09-22","releaseDate":"2025-09-22","access":"api","status":"previous","availability":"api","aliases":["deepseek-v3.1-terminus"],"tags":["上一代","推理","代码"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"DeepSeek V3.1 Terminus：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["deepseek-v3.1-terminus"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-deepseek-updates","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-deepseek-updates","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"deepseek/v3.2-exp@2025-09-29","canonicalId":"deepseek/v3.2-exp@2025-09-29","familyId":"deepseek/v3","name":"DeepSeek V3.2 Exp","provider":"DeepSeek","mark":"D","release":"2025-09-29","releaseDate":"2025-09-29","access":"open_weights","status":"previous","availability":"open_weights","aliases":["deepseek-v3.2-exp"],"tags":["上一代","开放权重","推理"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"DeepSeek V3.2 Exp：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["deepseek-v3.2-exp"],"availability":["open-weights"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-deepseek-updates","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-deepseek-updates","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"deepseek/v4-pro@2026-08-13","canonicalId":"deepseek/v4-pro@2026-08-13","familyId":"deepseek/v4","name":"DeepSeek V4 Pro (0813)","provider":"DeepSeek","mark":"D","release":"2026-08-13","releaseDate":"2026-08-13","access":"api","status":"active","availability":"api","aliases":["deepseek-v4-pro-0813"],"tags":["旗舰","推理","代码"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"DeepSeek V4 Pro (0813)：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["deepseek-v4-pro-0813"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-deepseek-v4","src-deepseek-updates"],"last_checked":"2026-08-27"},"sourceIds":["src-deepseek-v4","src-deepseek-updates"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"deepseek/v4-flash-vision-exp@current","canonicalId":"deepseek/v4-flash-vision-exp@current","familyId":"deepseek/v4","name":"DeepSeek V4 Flash Vision Exp","provider":"DeepSeek","mark":"D","release":null,"releaseDate":null,"access":"api","status":"preview","availability":"api","aliases":["deepseek-v4-flash-vision-exp"],"tags":["Flash","多模态","实验"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":true},"profile":{"positioning":"DeepSeek V4 Flash Vision Exp：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","多模态"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["deepseek-v4-flash-vision-exp"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-deepseek-v4","src-deepseek-updates"],"last_checked":"2026-08-27"},"sourceIds":["src-deepseek-v4","src-deepseek-updates"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"moonshot/kimi-k2.5@2026-01-27","canonicalId":"moonshot/kimi-k2.5@2026-01-27","familyId":"moonshot/kimi-k2","name":"Kimi K2.5","provider":"Moonshot AI","mark":"K","release":"2026-01-27","releaseDate":"2026-01-27","access":"api","status":"previous","availability":"api","aliases":["kimi-k2.5"],"tags":["上一代","多模态","Agent"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Kimi K2.5 上一代托管多模态/Agent endpoint，保留作 K2.6 与 K3 的代际基线。","capabilities":["推理","工具调用","多模态"],"best_for":["Kimi K2/K3 代际对照","托管多模态 Agent 报告追踪"],"endpoint_ids":["kimi-k2.5"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"该托管 endpoint 的参数量未公开。","license":null,"caveats":["Kimi Code、通用 API 与第三方 agent harness 是不同 system subject。","文本与视觉输入结果必须按 modality 和 endpoint 分列。"],"fact_source_ids":["src-kimi-code","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-kimi-code","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"zai/glm-4.7@2025-12-22","canonicalId":"zai/glm-4.7@2025-12-22","familyId":"zai/glm-4","name":"GLM-4.7","provider":"Z.ai","mark":"Z","release":"2025-12-22","releaseDate":"2025-12-22","access":"api","status":"previous","availability":"api","aliases":["glm-4.7"],"tags":["上一代","推理","代码"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GLM-4.7 上一代托管 reasoning/coding endpoint，保留作 GLM-5 系列之前的基线。","capabilities":["推理","代码"],"best_for":["GLM 4.x/5.x 代际对照","托管代码与 reasoning 报告追踪"],"endpoint_ids":["glm-4.7"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"该托管 endpoint 的参数量未公开。","license":null,"caveats":["代码 Agent 成绩必须保留工具、scaffold 与预算。","不得把 GLM-4.7 Flash 的服务结果合并到本条目。"],"fact_source_ids":["src-glm52","src-epoch-benchmark-hub"],"last_checked":"2026-08-28"},"sourceIds":["src-glm52","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"zai/glm-4.7-flash@current","canonicalId":"zai/glm-4.7-flash@current","familyId":"zai/glm-4","name":"GLM-4.7 Flash","provider":"Z.ai","mark":"Z","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["glm-4.7-flash"],"tags":["Flash","效率"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GLM-4.7 Flash 上一代托管效率 endpoint；Flash 只表示产品层级候选。","capabilities":["文本生成","推理"],"best_for":["GLM Flash 代际对照","质量/成本/延迟 Pareto 候选"],"endpoint_ids":["glm-4.7-flash"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"该托管 endpoint 的参数量未公开。","license":null,"caveats":["Flash 标签不能替代 TTFT、tokens/s、价格和 batch 条件。","不能把该 endpoint 的结果静默并入 GLM-4.7 主端点。"],"fact_source_ids":["src-glm52"],"last_checked":"2026-08-28"},"sourceIds":["src-glm52"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"minimax/m2.1@2025-12-23","canonicalId":"minimax/m2.1@2025-12-23","familyId":"minimax/m2","name":"MiniMax M2.1","provider":"MiniMax","mark":"M","release":"2025-12-23","releaseDate":"2025-12-23","access":"api","status":"previous","availability":"api","aliases":["minimax-m2.1"],"tags":["上一代","代码","Agent"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"MiniMax M2.1：公开榜单中有可追溯的 release/endpoint 记录。","capabilities":["推理","代码","工具调用"],"best_for":["代际 benchmark 对照","查看不同 effort/harness 下的公开报告"],"endpoint_ids":["minimax-m2.1"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"公开目录未统一核验；以每条结果的 protocol/context 记录为准。","parameter_note":"参数量未在本目录统一核验。","license":null,"caveats":["同一 release 的 effort、上下文、搜索和 quantization 展示保留在 source alias/raw modelRef；不可把它们当作独立复现。","跨来源比较前请检查 benchmark version、harness 和 subject type。"],"fact_source_ids":["src-minimax-release","src-epoch-benchmark-hub"],"last_checked":"2026-08-27"},"sourceIds":["src-minimax-release","src-epoch-benchmark-hub"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"xiaomi/mimo-v2-flash@current","canonicalId":"xiaomi/mimo-v2-flash@current","familyId":"xiaomi/mimo-v2","name":"MiMo V2 Flash","provider":"Xiaomi MiMo","mark":"X","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["mimo-v2-flash"],"tags":["Flash","效率","推理"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"MiMo V2 Flash 上一代托管效率 endpoint，用于 MiMo Flash/Fast 时间线。","capabilities":["文本生成","推理"],"best_for":["MiMo 效率端点代际对照","质量/成本/延迟 Pareto 候选"],"endpoint_ids":["mimo-v2-flash"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"该托管 endpoint 的参数量未公开。","license":null,"caveats":["Flash 标签不能替代 TTFT、tokens/s、价格和 batch 条件。","该 endpoint 不自动继承 MiMo V2 Pro 或 Omni 的能力与参数事实。"],"fact_source_ids":["src-mimo-models","src-mimo-updates"],"last_checked":"2026-08-28"},"sourceIds":["src-mimo-models","src-mimo-updates"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"xiaomi/mimo-v2-pro@current","canonicalId":"xiaomi/mimo-v2-pro@current","familyId":"xiaomi/mimo-v2","name":"MiMo V2 Pro","provider":"Xiaomi MiMo","mark":"X","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["mimo-v2-pro"],"tags":["推理","Agent"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"MiMo V2 Pro 上一代托管 reasoning/Agent endpoint，保留作 V2.5 Pro 的代际基线。","capabilities":["推理","工具调用"],"best_for":["MiMo Pro 代际对照","托管 Agent 与工具调用报告追踪"],"endpoint_ids":["mimo-v2-pro"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"该托管 endpoint 的参数量未公开。","license":null,"caveats":["Agent 成绩必须保留工具集、scaffold、预算与 endpoint。","不能从 V2.5 Pro 反向继承参数或 context。"],"fact_source_ids":["src-mimo-models","src-mimo-updates"],"last_checked":"2026-08-28"},"sourceIds":["src-mimo-models","src-mimo-updates"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"xiaomi/mimo-v2-omni@current","canonicalId":"xiaomi/mimo-v2-omni@current","familyId":"xiaomi/mimo-v2","name":"MiMo V2 Omni","provider":"Xiaomi MiMo","mark":"X","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["mimo-v2-omni"],"tags":["全模态","推理"],"summary":"来源榜单/模型卡明确出现；公开分数在页面标为披露·未复现。","modalities":["text","image","audio","video"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"MiMo V2 Omni 上一代全模态托管 endpoint，覆盖文本、图像、音频与视频输入。","capabilities":["推理","视觉输入","音频输入","视频输入"],"best_for":["MiMo Omni 代际对照","跨模态公开报告追踪"],"endpoint_ids":["mimo-v2-omni"],"availability":["api"],"reasoning_modes":["reasoning"],"architecture":null,"context_note":"该 endpoint 的 context 上限尚未在目录中逐项核验。","parameter_note":"该托管 endpoint 的参数量未公开。","license":null,"caveats":["不同输入模态的采样、费用和 grader 不可混为同一协议。","不能从 V2.5 Omni 类产品反向继承 context 或参数事实。"],"fact_source_ids":["src-mimo-models","src-mimo-updates"],"last_checked":"2026-08-28"},"sourceIds":["src-mimo-models","src-mimo-updates"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.4-pro@2026-03-05","canonicalId":"openai/gpt-5.4-pro@2026-03-05","familyId":"openai/gpt-5.4","name":"GPT-5.4 Pro","provider":"OpenAI","mark":"O","release":"2026-03-05","releaseDate":"2026-03-05","access":"api","status":"active","availability":"api","aliases":["gpt-5.4-pro","gpt-5.4-pro-2026-03-05"],"tags":["推理","高计算","专业工作"],"summary":"GPT-5.4 的 Pro 高计算 endpoint；官方规格记录 1,050,000-token context，参数量未披露。","modalities":["text","image"],"contextWindow":1050000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"tier":"pro","preview":false},"profile":{"positioning":"GPT-5.4 的 Pro 高计算 endpoint，面向低吞吐但高价值的复杂专业任务。","capabilities":["深度推理","代码","工具调用","视觉输入","专业知识工作"],"best_for":["复杂研究与审阅","高准确度代码任务","需要 Pro 计算预算的评测"],"endpoint_ids":["gpt-5.4-pro","gpt-5.4-pro-2026-03-05"],"availability":["api"],"reasoning_modes":["pro"],"architecture":null,"context_note":"OpenAI endpoint 规格记录 1,050,000-token context。","parameter_note":"参数量未公开。","knowledge_cutoff":"2025-08-31","max_output_tokens":128000,"license":null,"caveats":["Pro 是独立高计算 endpoint；不得把基础 GPT-5.4 的 protocol 或分数直接代入。","Epoch 的 none/unknown/xhigh/web-app 拼写保留为 source protocol，不是四个 release。"],"fact_source_ids":["src-openai-models","src-openai-gpt54-pro-spec"],"last_checked":"2026-08-28"},"sourceIds":["src-openai-models","src-openai-gpt54-pro-spec"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5.4-nano@2026-03-17","canonicalId":"openai/gpt-5.4-nano@2026-03-17","familyId":"openai/gpt-5.4","name":"GPT-5.4 nano","provider":"OpenAI","mark":"O","release":"2026-03-17","releaseDate":"2026-03-17","access":"api","status":"active","availability":"api","aliases":["gpt-5.4-nano","gpt-5.4-nano-2026-03-17"],"tags":["小模型","效率","多模态"],"summary":"GPT-5.4 nano dated snapshot；400K context、128K max output，支持 text/image input。","modalities":["text","image"],"contextWindow":400000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"tier":"nano","preview":false},"profile":{"positioning":"GPT-5.4 家族的 nano 效率 tier，提供大输入窗口与较长输出上限。","capabilities":["轻量推理","文本输入","图像输入","长上下文","结构化输出"],"best_for":["高吞吐 API 工作流","长输入抽取与分类","成本敏感的多模态任务"],"endpoint_ids":["gpt-5.4-nano","gpt-5.4-nano-2026-03-17"],"availability":["api"],"reasoning_modes":["none","low","medium","high","xhigh"],"architecture":null,"context_note":"官方目录记录 400K context。","parameter_note":"参数量未公开。","knowledge_cutoff":"2025-08-31","max_output_tokens":128000,"license":null,"caveats":["128K 是 max output，不应与 400K input context 相加后当作单一窗口。","LiveBench/Arena/Epoch 的 effort 后缀属于 run protocol。"],"fact_source_ids":["src-openai-models","src-openai-gpt54-nano-spec"],"last_checked":"2026-08-28"},"sourceIds":["src-openai-models","src-openai-gpt54-nano-spec"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5-mini@2025-08-07","canonicalId":"openai/gpt-5-mini@2025-08-07","familyId":"openai/gpt-5","name":"GPT-5 mini","provider":"OpenAI","mark":"O","release":"2025-08-07","releaseDate":"2025-08-07","access":"api","status":"previous","availability":"api","aliases":["gpt-5-mini","gpt-5-mini-2025-08-07"],"tags":["上一代","小模型","效率"],"summary":"GPT-5 的 dated mini release；未核验字段保持未知。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"tier":"mini","preview":false},"profile":{"positioning":"GPT-5 代际的 dated mini release，保留作成本/能力折中和历史公开榜单基线。","capabilities":["推理","代码","工具调用","视觉输入"],"best_for":["GPT-5 大小 tier 对照","成本敏感的通用任务","2025 年公开榜单回溯"],"endpoint_ids":["gpt-5-mini-2025-08-07"],"availability":["api"],"reasoning_modes":[],"architecture":null,"context_note":"该 dated mini release 的 context 上限未在本轮官方目录核验。","parameter_note":"参数量未公开；mini 是产品 tier，不是公开参数规模。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["BFCL、Epoch 和 Arena 的不同拼写只在各自 source scope 映射。","minimal/low/medium/high 后缀必须保留在 protocol。"],"fact_source_ids":["src-openai-models"],"last_checked":"2026-08-28"},"sourceIds":["src-openai-models"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"openai/gpt-5-nano@2025-08-07","canonicalId":"openai/gpt-5-nano@2025-08-07","familyId":"openai/gpt-5","name":"GPT-5 nano","provider":"OpenAI","mark":"O","release":"2025-08-07","releaseDate":"2025-08-07","access":"api","status":"previous","availability":"api","aliases":["gpt-5-nano","gpt-5-nano-2025-08-07"],"tags":["上一代","小模型","高吞吐"],"summary":"GPT-5 的 dated nano release；未核验字段保持未知。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"tier":"nano","preview":false},"profile":{"positioning":"GPT-5 代际的 dated nano release，面向最轻量、高吞吐的历史 API 比较。","capabilities":["轻量推理","代码辅助","工具调用","视觉输入"],"best_for":["批量分类与抽取","低成本路由","GPT-5 nano 历史评测复盘"],"endpoint_ids":["gpt-5-nano-2025-08-07"],"availability":["api"],"reasoning_modes":[],"architecture":null,"context_note":"该 dated nano release 的 context 上限未在本轮官方目录核验。","parameter_note":"参数量未公开；nano 标签不提供可推断的参数数值。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["不要从 GPT-5.4 nano 继承 400K/128K 限制。","Epoch effort 与 Arena high 是评测设置，不是不同模型。"],"fact_source_ids":["src-openai-models"],"last_checked":"2026-08-28"},"sourceIds":["src-openai-models"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.5-35b-a3b@current","canonicalId":"qwen/qwen3.5-35b-a3b@current","familyId":"qwen/qwen3.5","name":"Qwen3.5 35B-A3B","provider":"Qwen","mark":"Q","release":null,"releaseDate":null,"access":"open_weights","status":"previous","availability":"open_weights","aliases":["qwen3.5-35b-a3b","Qwen/Qwen3.5-35B-A3B"],"tags":["MoE","开放权重","多模态","效率"],"summary":"35B total / 3B active，原生 262,144-token context，Apache-2.0 开放权重。","modalities":["text","image"],"contextWindow":262144,"paramsTotal":35000000000,"paramsActive":3000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.5 的 35B-A3B 多模态 MoE 开放权重版本，强调低 active-parameter 部署效率。","capabilities":["文本","图像","推理","代码","多语言"],"best_for":["3B active 的本地/私有部署","多模态开源模型比较","Qwen3.5 小型 MoE 研究"],"endpoint_ids":["Qwen/Qwen3.5-35B-A3B"],"availability":["open-weights"],"reasoning_modes":[],"architecture":"MoE","context_note":"原生 262,144-token context。","parameter_note":"35B total / 3B active。","knowledge_cutoff":null,"max_output_tokens":null,"license":"Apache-2.0","caveats":["total 与 active 参数不可互换。","本地量化与视觉预处理属于 deployment/protocol，而非新 release。"],"fact_source_ids":["src-qwen35"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen35"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.5-9b@current","canonicalId":"qwen/qwen3.5-9b@current","familyId":"qwen/qwen3.5","name":"Qwen3.5 9B","provider":"Qwen","mark":"Q","release":null,"releaseDate":null,"access":"open_weights","status":"previous","availability":"open_weights","aliases":["qwen3.5-9b","Qwen/Qwen3.5-9B"],"tags":["Dense","开放权重","多模态","小模型"],"summary":"9B dense 开放权重多模态模型；未确定的 context 等字段保留为未知。","modalities":["text","image"],"contextWindow":null,"paramsTotal":9000000000,"paramsActive":9000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Qwen3.5 9B dense 多模态开放权重模型，作为轻量 dense 基线。","capabilities":["文本","图像","代码","中文/多语言"],"best_for":["单机部署","dense 与 MoE 对照","轻量多模态实验"],"endpoint_ids":["Qwen/Qwen3.5-9B"],"availability":["open-weights"],"reasoning_modes":[],"architecture":"Dense","context_note":"本轮来源未确认可稳定复用的 context 数值。","parameter_note":"9B dense；按 dense 口径 total 与 active 均为 9B。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["未确认字段保持 null，不从相邻 Qwen3.5 checkpoint 继承。","视觉 benchmark 需记录分辨率和预处理。"],"fact_source_ids":["src-qwen35"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen35"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3.5-max-preview@2026-03-18","canonicalId":"qwen/qwen3.5-max-preview@2026-03-18","familyId":"qwen/qwen3.5","name":"Qwen3.5 Max Preview","provider":"Qwen","mark":"Q","release":"2026-03-18","releaseDate":"2026-03-18","access":"api","status":"preview","availability":"api","aliases":["qwen3.5-max-preview"],"tags":["preview","托管服务","旗舰"],"summary":"Qwen3.5 hosted preview endpoint；参数规模未知，不与开放权重 checkpoint 自动合并。","modalities":["text"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":true},"profile":{"positioning":"Qwen3.5 Max 的 hosted preview endpoint，用于观察托管旗舰迭代而非开放权重部署。","capabilities":["推理","代码","中文/多语言","托管服务"],"best_for":["Qwen hosted Max 时间线","preview 公开报告审计","服务端旗舰对照"],"endpoint_ids":["qwen3.5-max-preview"],"availability":["hosted-preview","api"],"reasoning_modes":[],"architecture":null,"context_note":"Preview endpoint 的 context 上限未确认。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["Preview 路由可变化，结果必须保留 observed date。","不得与 35B/9B 开放权重条目按名称自动合并。"],"fact_source_ids":["src-qwen35-max-preview"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen35-max-preview"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"qwen/qwen3-coder-next@2026-02-03","canonicalId":"qwen/qwen3-coder-next@2026-02-03","familyId":"qwen/qwen3-coder","name":"Qwen3-Coder-Next","provider":"Qwen","mark":"Q","release":"2026-02-03","releaseDate":"2026-02-03","access":"open_weights","status":"active","availability":"open_weights","aliases":["qwen3-coder-next","Qwen/Qwen3-Coder-Next"],"tags":["代码","MoE","开放权重","non-thinking"],"summary":"80B total / 3B active、原生 262,144-token context 的 non-thinking coding model，Apache-2.0。","modalities":["text"],"contextWindow":262144,"paramsTotal":80000000000,"paramsActive":3000000000,"paramsApproximate":false,"openWeights":true,"variant":{"reasoning":false,"preview":false},"profile":{"positioning":"Qwen3-Coder-Next 80B-A3B non-thinking 开放权重代码模型，偏向高效软件工程推理。","capabilities":["代码生成","仓库理解","工具调用","代码 Agent"],"best_for":["SWE-bench 类任务","本地代码 agent","3B active 的高吞吐 coding"],"endpoint_ids":["Qwen/Qwen3-Coder-Next","qwen3-coder-next"],"availability":["open-weights"],"reasoning_modes":["non-thinking"],"architecture":"MoE","context_note":"原生 262,144-token context。","parameter_note":"80B total / 3B active。","knowledge_cutoff":null,"max_output_tokens":null,"license":"Apache-2.0","caveats":["non-thinking 是该 release 的工作模式，不应用 thinking 分数补齐。","SWE/Terminal 成绩必须保留 harness、工具、预算和 checkpoint。"],"fact_source_ids":["src-qwen3-coder-next"],"last_checked":"2026-08-28"},"sourceIds":["src-qwen3-coder-next"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"google/gemini-3.7-flash@current","canonicalId":"google/gemini-3.7-flash@current","familyId":"google/gemini-3.7","name":"Gemini 3.7 Flash","provider":"Google","mark":"G","release":null,"releaseDate":null,"access":"api","status":"active","availability":"api","aliases":["gemini-3.7-flash"],"tags":["Flash","最新","多模态","效率"],"summary":"Google 官方目录中的当前 latest Flash endpoint；1,048,576 input、65,536 max output，参数规模未披露。","modalities":["text","image"],"contextWindow":1048576,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"speed_variant":true},"profile":{"positioning":"Google 官方目录中的 latest Flash endpoint，代表当前 Gemini 快速通用 tier。","capabilities":["快速推理","文本","图像","工具型工作流"],"best_for":["当前 Flash 横向比较","高吞吐多模态应用","低延迟交互"],"endpoint_ids":["gemini-3.7-flash"],"availability":["api","hosted"],"reasoning_modes":[],"architecture":null,"context_note":"官方 endpoint 规格记录 1,048,576 input tokens。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":65536,"license":null,"caveats":["latest 表示 catalog 生命周期，不授权映射 floating latest alias。","high/medium/minimal 等后缀属于来源 protocol。"],"fact_source_ids":["src-google-gemini-models","src-google-gemini37-spec"],"last_checked":"2026-08-28"},"sourceIds":["src-google-gemini-models","src-google-gemini37-spec"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"google/gemini-3.6-flash@current","canonicalId":"google/gemini-3.6-flash@current","familyId":"google/gemini-3.6","name":"Gemini 3.6 Flash","provider":"Google","mark":"G","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["gemini-3.6-flash"],"tags":["Flash","上一代","多模态"],"summary":"Google 官方目录中的 previous-generation Flash endpoint；context 与参数规模未知。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"speed_variant":true},"profile":{"positioning":"Google 官方目录标注的 previous-generation Flash，保留作 3.7 的直接代际基线。","capabilities":["快速推理","文本","图像","多模态交互"],"best_for":["Gemini 3.7/3.6 代际比较","历史 Arena/LiveBench 回溯","Flash 服务迁移评估"],"endpoint_ids":["gemini-3.6-flash"],"availability":["api","hosted"],"reasoning_modes":[],"architecture":null,"context_note":"本轮官方 catalog 未给出可登记的 context 数值。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["previous-generation 不等于 deprecated。","Arena/LiveBench 的 high 展示不得与默认 endpoint 分数静默合并。"],"fact_source_ids":["src-google-gemini-models"],"last_checked":"2026-08-28"},"sourceIds":["src-google-gemini-models"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"google/gemini-3.5-flash@current","canonicalId":"google/gemini-3.5-flash@current","familyId":"google/gemini-3.5","name":"Gemini 3.5 Flash","provider":"Google","mark":"G","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["gemini-3.5-flash"],"tags":["Flash","legacy","多模态"],"summary":"Google 官方目录中的 legacy Flash endpoint；保留作历史比较。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"speed_variant":true},"profile":{"positioning":"Google 官方目录中的 legacy Gemini 3.5 Flash，服务于旧报告和生命周期追踪。","capabilities":["文本","图像","快速生成","工具型工作流"],"best_for":["legacy benchmark 回放","3.5→3.6→3.7 代际趋势","旧 endpoint 迁移审计"],"endpoint_ids":["gemini-3.5-flash"],"availability":["api","hosted"],"reasoning_modes":[],"architecture":null,"context_note":"Legacy catalog 条目的 context 数值未在本轮核验。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["legacy 是目录定位，不代表端点已正式 deprecated。","不同 effort 后缀仍需保留原 source name。"],"fact_source_ids":["src-google-gemini-models"],"last_checked":"2026-08-28"},"sourceIds":["src-google-gemini-models"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"google/gemini-3.5-flash-lite@current","canonicalId":"google/gemini-3.5-flash-lite@current","familyId":"google/gemini-3.5","name":"Gemini 3.5 Flash-Lite","provider":"Google","mark":"G","release":null,"releaseDate":null,"access":"api","status":"previous","availability":"api","aliases":["gemini-3.5-flash-lite"],"tags":["Flash-Lite","效率","legacy"],"summary":"Gemini 3.5 系列的 efficiency-oriented Flash-Lite endpoint。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false,"speed_variant":true},"profile":{"positioning":"Gemini 3.5 家族的 efficiency-oriented Flash-Lite endpoint，面向更低成本和延迟。","capabilities":["轻量生成","文本","图像","高吞吐"],"best_for":["Flash-Lite 效率比较","批处理与分类","成本敏感多模态请求"],"endpoint_ids":["gemini-3.5-flash-lite"],"availability":["api","hosted"],"reasoning_modes":[],"architecture":null,"context_note":"本轮官方 catalog 未确认 context 数值。","parameter_note":"参数量未公开；Lite 不能换算为参数规模。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["Flash-Lite 与 Flash 是不同 endpoint，不共享分数。","效率比较仍需记录价格、延迟和 effort protocol。"],"fact_source_ids":["src-google-gemini-models"],"last_checked":"2026-08-28"},"sourceIds":["src-google-gemini-models"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"xai/grok-4.6@2026-08-12","canonicalId":"xai/grok-4.6@2026-08-12","familyId":"xai/grok-4","name":"Grok 4.6","provider":"xAI","mark":"X","release":"2026-08-12","releaseDate":"2026-08-12","access":"api","status":"active","availability":"api","aliases":["grok-4.6"],"tags":["旗舰","推理","多模态"],"summary":"xAI 当前 Grok 旗舰；官方 API 规格披露 500K context，参数规模未披露。","modalities":["text","image"],"contextWindow":500000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"xAI 当前 Grok 旗舰，用于最新 reasoning、代码和工具型公开报告。","capabilities":["推理","代码","视觉输入","工具型工作流"],"best_for":["当前 Grok frontier 比较","高 effort 推理报告","Grok 4.5→4.6 迁移"],"endpoint_ids":["grok-4.6"],"availability":["api","hosted"],"reasoning_modes":[],"architecture":"Native text-and-image input, text output（capability-level disclosure）","context_note":"xAI API 规格记录 500,000-token context。","parameter_note":"参数量未公开。","knowledge_cutoff":"2026-01","max_output_tokens":null,"license":null,"caveats":["xhigh/high/medium/low 后缀是 source protocol，不是独立 release。","带检索或工具的 system 结果必须保留 harness。"],"fact_source_ids":["src-xai-grok46","src-xai-grok46-spec"],"last_checked":"2026-08-28"},"sourceIds":["src-xai-grok46","src-xai-grok46-spec"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"xai/grok-4.5@2026-07-16","canonicalId":"xai/grok-4.5@2026-07-16","familyId":"xai/grok-4","name":"Grok 4.5","provider":"xAI","mark":"X","release":"2026-07-16","releaseDate":"2026-07-16","access":"api","status":"previous","availability":"api","aliases":["grok-4.5"],"tags":["上一代","推理","多模态"],"summary":"Grok 4.6 的 previous-generation baseline；参数规模未知。","modalities":["text","image"],"contextWindow":null,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"Grok 4.6 的 previous-generation 旗舰基线，用于近期代际与 effort 对照。","capabilities":["推理","代码","视觉输入","工具型工作流"],"best_for":["Grok 代际时间线","2026 年中公开榜单复盘","reasoning effort 灵敏度比较"],"endpoint_ids":["grok-4.5"],"availability":["api","hosted"],"reasoning_modes":[],"architecture":null,"context_note":"当前来源未公开可登记的 context 数值。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":null,"license":null,"caveats":["previous 不表示已退役。","Arena、LiveBench、Epoch 与 Agents' Last Exam 的拼写只在各自 source scope 映射。"],"fact_source_ids":["src-xai-grok45"],"last_checked":"2026-08-28"},"sourceIds":["src-xai-grok45"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0},{"id":"zai/glm-5v-turbo@2026-04-01","canonicalId":"zai/glm-5v-turbo@2026-04-01","familyId":"zai/glm-5v","name":"GLM-5V-Turbo","provider":"Z.ai","mark":"Z","release":"2026-04-01","releaseDate":"2026-04-01","access":"api","status":"active","availability":"api","aliases":["glm-5v-turbo"],"tags":["多模态","代码","视觉","Agent"],"summary":"面向多模态 coding 的 API model；text/image/video/file input、text output、200K context、128K max output。","modalities":["text","image","video","file"],"contextWindow":200000,"paramsTotal":null,"paramsActive":null,"paramsApproximate":false,"openWeights":false,"variant":{"reasoning":true,"preview":false},"profile":{"positioning":"GLM-5V-Turbo 多模态 coding API model，统一处理 text/image/video/file 输入并输出文本。","capabilities":["文本输入","图像理解","视频理解","文件理解","多模态代码生成"],"best_for":["截图/视频到代码","多文件视觉分析","长上下文多模态 coding agent"],"endpoint_ids":["glm-5v-turbo"],"availability":["api","hosted"],"reasoning_modes":[],"architecture":null,"context_note":"200K context；text output 上限 128K tokens。","parameter_note":"参数量未公开。","knowledge_cutoff":null,"max_output_tokens":128000,"license":null,"caveats":["file 是输入载体而非输出模态；输出为 text。","多模态 coding 结果依赖文件解析、采样、工具和视觉预处理。"],"fact_source_ids":["src-glm5v-turbo"],"last_checked":"2026-08-28"},"sourceIds":["src-glm5v-turbo"],"scores":{},"scoreCount":0,"catalogOnly":true,"systemRunCount":0}],"harnesses":[{"id":"model-only","name":"Model-only / single turn","label":"model-only","kind":"model","status":"active","owner":"benchmark protocol","description":"无外部 agent loop；保留 prompt、shots、解码和 reasoning 设置。","tools":[],"source_ids":[]},{"id":"unspecified-reported","name":"Provider-reported (harness unspecified)","label":"未注明 harness","kind":"unknown","status":"active","owner":"provider report","description":"来源报告了结果但没有足够信息确定执行器；不得与固定 harness 的结果做 exact 比较。","tools":null,"source_ids":[]},{"id":"swebench-official","name":"SWE-bench official harness","label":"SWE-bench official","kind":"benchmark","status":"active","owner":"SWE-bench","description":"官方 SWE-bench evaluator；实例、补丁应用和测试环境按 benchmark 版本固定。","tools":["repository","shell","patch"],"source_ids":["swebench"]},{"id":"mini-swe-agent","name":"mini-SWE-agent","label":"mini-SWE-agent","kind":"coding-agent","status":"active","owner":"SWE-bench community","description":"轻量 coding agent scaffold；记录版本、timeout、temperature 和 rollout 聚合。","tools":["shell","editor"],"source_ids":["src-deepswe"]},{"id":"terminus-2","name":"Terminus-2","label":"Terminus-2","kind":"terminal-agent","status":"active","owner":"Terminal-Bench","description":"Terminal-Bench 常见终端 agent；不同版本和 effort 是不同 protocol。","tools":["terminal","filesystem"],"source_ids":["terminal-bench"]},{"id":"claude-code","name":"Claude Code","label":"Claude Code","kind":"coding-agent","status":"active","owner":"Anthropic","description":"Claude Code CLI/agent scaffold；记录版本、权限、工具和上下文管理。","tools":["shell","filesystem","editor"],"source_ids":["src-anthropic-system-cards"]},{"id":"codex","name":"Codex harness","label":"Codex","kind":"coding-agent","status":"active","owner":"OpenAI","description":"Codex/Responses coding agent scaffold；应记录 effort、sandbox、tools 和 timeout。","tools":["shell","filesystem","editor"],"source_ids":["src-openai-gpt56"]},{"id":"kimi-code","name":"Kimi Code","label":"Kimi Code","kind":"coding-agent","status":"active","owner":"Moonshot AI","description":"Kimi 官方 coding harness；K3 官方 benchmark 明确使用该 harness。","tools":["shell","filesystem","editor"],"source_ids":["src-kimi-k3","src-kimi-code"]},{"id":"deepseek-harness","name":"DeepSeek Harness","label":"DeepSeek Harness","kind":"coding-agent","status":"active","owner":"DeepSeek","description":"DeepSeek V4 Flash 官方 agent 评测 harness；官方更新注明 minimal mode/max effort。","tools":["shell","filesystem","editor"],"source_ids":["src-deepseek-updates"]},{"id":"lm-eval","name":"EleutherAI lm-evaluation-harness","label":"lm-eval","kind":"static-eval","status":"active","owner":"EleutherAI","description":"静态 benchmark runner；记录 commit、task config、shots 和 generation 参数。","tools":[],"source_ids":["src-helm"]},{"id":"vendor-default","name":"Vendor default evaluator","label":"厂商默认","kind":"provider","status":"active","owner":"provider","description":"厂商模型卡中的默认评测设置；作为 reported/conditional，不推断具体 agent。","tools":null,"source_ids":[]},{"id":"arena-human-preference","name":"Arena human-preference rating","label":"Arena human preference","kind":"preference-eval","status":"active","owner":"Arena / LMArena","description":"Arena pairwise human-vote aggregation; rating, CI, votes and category stay attached to each snapshot.","tools":["human-vote"],"source_ids":["lmarena-hf-dataset"]},{"id":"arena-agent","name":"Arena Agent IPS evaluator","label":"Arena Agent IPS","kind":"agent-eval","status":"active","owner":"Arena / LMArena","description":"Arena Agent published IPS (tau-hat) score; not interchangeable with Bradley-Terry Elo or model-only runs.","tools":["agent-session","human-signal"],"source_ids":["lmarena-hf-dataset"]},{"id":"agents-last-exam","name":"Agents' Last Exam evaluator","label":"ALE","kind":"agent-eval","status":"active","owner":"UC Berkeley RDI","description":"ALE-V1 sandbox/evaluator; source agent harness (Codex, Claude Code, Kimi Code, ALE-Claw and others) and effort variant remain in each run's protocol.","tools":["sandbox","desktop","terminal","grader"],"source_ids":["agents-last-exam"]},{"id":"bfcl-eval","name":"Berkeley Function Calling evaluator","label":"BFCL evaluator","kind":"tool-use-eval","status":"active","owner":"Berkeley Gorilla Lab","description":"BFCL V4 evaluator；必须区分 native function calling（FC）与 prompt workaround，并记录固定 evaluator commit。","tools":["function-calling","prompt","tool-schema"],"source_ids":["src-bfcl"]}],"runs":[{"id":"obs-qwen38-flash-swepro-20260826","modelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","provider":"Qwen","familyId":"qwen/qwen3.8","endpointId":"qwen3.8-flash","harnessId":"unspecified-reported","harnessName":"Provider-reported (harness unspecified)","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersion":"swebench-pro@v1","metricId":"resolved","metric":"Resolved","value":62.5,"rawValue":"62.5%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":true,"reasoning_mode":"reported","raw_setting":"SWE-bench Pro; provider reported"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":"provider coding-agent pipeline"},"subjectType":"system","evidence":{"level":"B","sourceIds":["src-qwen38-flash"],"evidenceIds":[],"sourceUrl":"https://qwen.ai/blog?id=qwen3.8-flash-next","sourceLabel":"Qwen · Qwen3.8-Flash-Next"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-qwen38-flash","sourceUrl":"https://qwen.ai/blog?id=qwen3.8-flash-next","observedAt":"2026-08-26","publishedAt":"2026-08-26","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"System score; the source does not identify a canonical harness, so it remains conditional.","legacyMigrated":false},{"id":"obs-qwen38-flash-lcb-20260826","modelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","provider":"Qwen","familyId":"qwen/qwen3.8","endpointId":"qwen3.8-flash","harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"livecodebench","benchmarkName":"LiveCodeBench","benchmarkVersion":"livecodebench@v6","metricId":"pass@1","metric":"Pass@1","value":91.9,"rawValue":"91.9%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"LiveCodeBench v6; Qwen official table"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["src-qwen38-flash"],"evidenceIds":[],"sourceUrl":"https://qwen.ai/blog?id=qwen3.8-flash-next","sourceLabel":"Qwen · Qwen3.8-Flash-Next"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-qwen38-flash","sourceUrl":"https://qwen.ai/blog?id=qwen3.8-flash-next","observedAt":"2026-08-26","publishedAt":"2026-08-26","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Provider-reported rolling benchmark value; window and evaluator should be checked before exact comparison.","legacyMigrated":false},{"id":"obs-qwen38-flash-gpqa-20260826","modelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","provider":"Qwen","familyId":"qwen/qwen3.8","endpointId":"qwen3.8-flash","harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"gpqa-diamond@v1","metricId":"accuracy","metric":"准确率","value":91.7,"rawValue":"91.7%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"Qwen3.8-Flash-Next official table"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["src-qwen38-flash"],"evidenceIds":[],"sourceUrl":"https://qwen.ai/blog?id=qwen3.8-flash-next","sourceLabel":"Qwen · Qwen3.8-Flash-Next"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-qwen38-flash","sourceUrl":"https://qwen.ai/blog?id=qwen3.8-flash-next","observedAt":"2026-08-26","publishedAt":"2026-08-26","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Provider-reported value; protocol details follow the linked Qwen release page.","legacyMigrated":false},{"id":"obs-glm53-terminal21-20260814","modelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","provider":"Z.ai","familyId":"zai/glm-5","endpointId":"glm-5.3","harnessId":"claude-code","harnessName":"Claude Code","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersion":"terminal-bench@2.1","metricId":"pass_rate","metric":"Pass rate","value":88.2,"rawValue":"88.2%","unit":"%","protocol":{"shots":null,"temperature":1.0,"top_p":1.0,"tools":true,"reasoning_mode":"max","max_tokens":65536,"raw_setting":"Claude Code 2.1.207; 6h timeout"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":"Claude Code 2.1.207"},"subjectType":"system","evidence":{"level":"B","sourceIds":["src-glm53"],"evidenceIds":[],"sourceUrl":"https://z.ai/blog/glm-5.3","sourceLabel":"Z.ai · GLM-5.3"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-glm53","sourceUrl":"https://z.ai/blog/glm-5.3","observedAt":"2026-08-14","publishedAt":"2026-08-14","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Official GLM-5.3 comparison table and footnotes.","legacyMigrated":false},{"id":"obs-glm53-deepswe-20260814","modelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","provider":"Z.ai","familyId":"zai/glm-5","endpointId":"glm-5.3","harnessId":"mini-swe-agent","harnessName":"mini-SWE-agent","benchmarkId":"deepswe","benchmarkName":"DeepSWE v1.1","benchmarkVersion":"deepswe@v1.1","metricId":"resolved","metric":"Resolved","value":66.9,"rawValue":"66.9%","unit":"%","protocol":{"shots":null,"temperature":0.95,"top_p":1.0,"tools":true,"reasoning_mode":"max","max_tokens":null,"raw_setting":"DeepSWE v1.1; mini-SWE-agent; 6h timeout"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":"mini-SWE-agent"},"subjectType":"system","evidence":{"level":"B","sourceIds":["src-glm53"],"evidenceIds":[],"sourceUrl":"https://z.ai/blog/glm-5.3","sourceLabel":"Z.ai · GLM-5.3"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-glm53","sourceUrl":"https://z.ai/blog/glm-5.3","observedAt":"2026-08-14","publishedAt":"2026-08-14","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Official GLM-5.3 comparison table and footnotes.","legacyMigrated":false},{"id":"obs-glm53-cybergym-20260814","modelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","provider":"Z.ai","familyId":"zai/glm-5","endpointId":"glm-5.3","harnessId":"claude-code","harnessName":"Claude Code","benchmarkId":"cybergym","benchmarkName":"CyberGym","benchmarkVersion":"cybergym@v1","metricId":"success","metric":"Success","value":84.5,"rawValue":"84.5%","unit":"%","protocol":{"shots":null,"temperature":1.0,"top_p":1.0,"tools":true,"reasoning_mode":"max","max_tokens":128000,"raw_setting":"CyberGym; no web tools; 300 interaction rounds"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":"Claude Code 2.1.207"},"subjectType":"system","evidence":{"level":"B","sourceIds":["src-glm53"],"evidenceIds":[],"sourceUrl":"https://z.ai/blog/glm-5.3","sourceLabel":"Z.ai · GLM-5.3"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-glm53","sourceUrl":"https://z.ai/blog/glm-5.3","observedAt":"2026-08-14","publishedAt":"2026-08-14","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Official provider-reported security benchmark; keep separate from general coding scores.","legacyMigrated":false},{"id":"obs-deepseek-v4-flash-toolathlon-20260731","modelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","provider":"DeepSeek","familyId":"deepseek/v4","endpointId":"deepseek-v4-flash","harnessId":"deepseek-harness","harnessName":"DeepSeek Harness","benchmarkId":"toolathlon","benchmarkName":"Toolathlon","benchmarkVersion":"toolathlon@v1","metricId":"success","metric":"Success","value":70.3,"rawValue":"70.3%","unit":"%","protocol":{"shots":null,"temperature":1.0,"top_p":0.95,"tools":true,"reasoning_mode":"max","max_tokens":null,"raw_setting":"Toolathlon verified; DeepSeek Harness minimal mode"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":"DeepSeek Code Agent"},"subjectType":"system","evidence":{"level":"B","sourceIds":["src-deepseek-updates"],"evidenceIds":[],"sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API change log"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-deepseek-updates","sourceUrl":"https://api-docs.deepseek.com/updates/","observedAt":"2026-07-31","publishedAt":"2026-07-31","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Official update; verified subset as named by provider.","legacyMigrated":false},{"id":"obs-deepseek-v4-flash-terminal21-20260731","modelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","provider":"DeepSeek","familyId":"deepseek/v4","endpointId":"deepseek-v4-flash","harnessId":"deepseek-harness","harnessName":"DeepSeek Harness","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersion":"terminal-bench@2.1","metricId":"pass_rate","metric":"Pass rate","value":82.7,"rawValue":"82.7%","unit":"%","protocol":{"shots":null,"temperature":1.0,"top_p":0.95,"tools":true,"reasoning_mode":"max","max_tokens":null,"raw_setting":"DeepSeek Harness minimal mode; max effort"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":"DeepSeek Code Agent"},"subjectType":"system","evidence":{"level":"B","sourceIds":["src-deepseek-updates"],"evidenceIds":[],"sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API change log"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-deepseek-updates","sourceUrl":"https://api-docs.deepseek.com/updates/","observedAt":"2026-07-31","publishedAt":"2026-07-31","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Official update; this is a system score, not a model-only score.","legacyMigrated":false},{"id":"obs-deepseek-v4-flash-nl2repo-20260731","modelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","provider":"DeepSeek","familyId":"deepseek/v4","endpointId":"deepseek-v4-flash","harnessId":"deepseek-harness","harnessName":"DeepSeek Harness","benchmarkId":"nl2repo","benchmarkName":"NL2Repo","benchmarkVersion":"nl2repo@v1","metricId":"resolved","metric":"Resolved","value":54.2,"rawValue":"54.2%","unit":"%","protocol":{"shots":null,"temperature":1.0,"top_p":0.95,"tools":true,"reasoning_mode":"max","max_tokens":null,"raw_setting":"DeepSeek Harness minimal mode; max effort"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":"DeepSeek Code Agent"},"subjectType":"system","evidence":{"level":"B","sourceIds":["src-deepseek-updates"],"evidenceIds":[],"sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API change log"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-deepseek-updates","sourceUrl":"https://api-docs.deepseek.com/updates/","observedAt":"2026-07-31","publishedAt":"2026-07-31","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Official update; public benchmark result.","legacyMigrated":false},{"id":"obs-deepseek-v4-flash-deepswe-20260731","modelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","provider":"DeepSeek","familyId":"deepseek/v4","endpointId":"deepseek-v4-flash","harnessId":"deepseek-harness","harnessName":"DeepSeek Harness","benchmarkId":"deepswe","benchmarkName":"DeepSWE v1.1","benchmarkVersion":"deepswe@v1.1","metricId":"resolved","metric":"Resolved","value":54.4,"rawValue":"54.4%","unit":"%","protocol":{"shots":null,"temperature":1.0,"top_p":0.95,"tools":true,"reasoning_mode":"max","max_tokens":null,"raw_setting":"DeepSeek Harness minimal mode; max effort"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":"DeepSeek Code Agent"},"subjectType":"system","evidence":{"level":"B","sourceIds":["src-deepseek-updates"],"evidenceIds":[],"sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API change log"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-deepseek-updates","sourceUrl":"https://api-docs.deepseek.com/updates/","observedAt":"2026-07-31","publishedAt":"2026-07-31","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Official update; public benchmark result.","legacyMigrated":false},{"id":"obs-kimi-k3-terminal21-20260716","modelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","provider":"Moonshot AI","familyId":"moonshot/kimi-k3","endpointId":"k3","harnessId":"kimi-code","harnessName":"Kimi Code","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersion":"terminal-bench@2.1","metricId":"pass_rate","metric":"Pass rate","value":88.3,"rawValue":"88.3%","unit":"%","protocol":{"shots":null,"temperature":1.0,"top_p":1.0,"tools":true,"reasoning_mode":"max","max_tokens":null,"raw_setting":"Kimi K3 official table; Kimi Code harness"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":"Kimi Code"},"subjectType":"system","evidence":{"level":"B","sourceIds":["src-kimi-k3"],"evidenceIds":[],"sourceUrl":"https://www.kimi.com/en/blog/kimi-k3","sourceLabel":"Kimi · K3 technical blog"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-kimi-k3","sourceUrl":"https://www.kimi.com/en/blog/kimi-k3","observedAt":"2026-07-16","publishedAt":"2026-07-16","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Kimi official page states that K3 uses Kimi Code while comparison models may use different harnesses.","legacyMigrated":false},{"id":"obs-kimi-k3-deepswe-20260716","modelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","provider":"Moonshot AI","familyId":"moonshot/kimi-k3","endpointId":"k3","harnessId":"kimi-code","harnessName":"Kimi Code","benchmarkId":"deepswe","benchmarkName":"DeepSWE v1.1","benchmarkVersion":"deepswe@v1.1","metricId":"resolved","metric":"Resolved","value":67.3,"rawValue":"67.3%","unit":"%","protocol":{"shots":null,"temperature":1.0,"top_p":1.0,"tools":true,"reasoning_mode":"max","max_tokens":null,"raw_setting":"DeepSWE v1.1; Kimi Code evaluation"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":"Kimi Code"},"subjectType":"system","evidence":{"level":"B","sourceIds":["src-kimi-k3"],"evidenceIds":[],"sourceUrl":"https://www.kimi.com/en/blog/kimi-k3","sourceLabel":"Kimi · K3 technical blog"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-kimi-k3","sourceUrl":"https://www.kimi.com/en/blog/kimi-k3","observedAt":"2026-07-16","publishedAt":"2026-07-16","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"The page notes the leaderboard baseline and harness differences; retain as a conditional system observation.","legacyMigrated":false},{"id":"obs-gpt56-sol-mmmupro-tools-20260827","modelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","provider":"OpenAI","familyId":"openai/gpt-5.6","endpointId":"gpt-5.6-sol","harnessId":"unspecified-reported","harnessName":"Provider-reported (harness unspecified)","benchmarkId":"mmmu-pro","benchmarkName":"MMMU-Pro","benchmarkVersion":"mmmu-pro@v1","metricId":"accuracy","metric":"准确率","value":84.6,"rawValue":"84.6%","unit":"%","protocol":{"shots":0,"temperature":null,"top_p":null,"tools":true,"reasoning_mode":"reported","raw_setting":"MMMU-Pro; with tools"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":"provider tools"},"subjectType":"system","evidence":{"level":"B","sourceIds":["src-openai-gpt56"],"evidenceIds":[],"sourceUrl":"https://openai.com/index/gpt-5-6/","sourceLabel":"OpenAI · GPT-5.6 launch and evaluations"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-openai-gpt56","sourceUrl":"https://openai.com/index/gpt-5-6/","observedAt":"2026-07-09","publishedAt":"2026-07-09","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":false,"notes":"Tool-enabled provider result; harness identity is not specified in the source.","legacyMigrated":false},{"id":"obs-gpt56-sol-mmmupro-no-tools-20260827","modelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","provider":"OpenAI","familyId":"openai/gpt-5.6","endpointId":"gpt-5.6-sol","harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"mmmu-pro","benchmarkName":"MMMU-Pro","benchmarkVersion":"mmmu-pro@v1","metricId":"accuracy","metric":"准确率","value":83.0,"rawValue":"83.0%","unit":"%","protocol":{"shots":0,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"MMMU-Pro; no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["src-openai-gpt56"],"evidenceIds":[],"sourceUrl":"https://openai.com/index/gpt-5-6/","sourceLabel":"OpenAI · GPT-5.6 launch and evaluations"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-openai-gpt56","sourceUrl":"https://openai.com/index/gpt-5-6/","observedAt":"2026-07-09","publishedAt":"2026-07-09","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"No-tools variant is preferred for the model atlas; the with-tools variant is retained as a separate run.","legacyMigrated":false},{"id":"obs-gpt56-sol-gpqa-20260827","modelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","provider":"OpenAI","familyId":"openai/gpt-5.6","endpointId":"gpt-5.6-sol","harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"gpqa-diamond@v1","metricId":"accuracy","metric":"准确率","value":94.6,"rawValue":"94.6%","unit":"%","protocol":{"shots":0,"temperature":null,"top_p":null,"tools":false,"reasoning_mode":"reported","raw_setting":"official GPT-5.6 evaluation table; no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["src-openai-gpt56"],"evidenceIds":[],"sourceUrl":"https://openai.com/index/gpt-5-6/","sourceLabel":"OpenAI · GPT-5.6 launch and evaluations"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"src-openai-gpt56","sourceUrl":"https://openai.com/index/gpt-5-6/","observedAt":"2026-07-09","publishedAt":"2026-07-09","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Provider-reported value; exact reasoning budget and evaluator details remain in the linked launch table.","legacyMigrated":false},{"id":"legacy-gemini-3-1-pro-terminal-bench","modelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","provider":"Google","familyId":"google/gemini-3.1","endpointId":null,"harnessId":"unspecified-reported","harnessName":"Provider-reported (harness unspecified)","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersion":"terminal-bench@2.0","metricId":"pass_rate","metric":"Pass rate","value":68.5,"rawValue":"68.5%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"Terminal-Bench 2.0 · Terminus-2"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"system","evidence":{"level":"B","sourceIds":["google-gemini31"],"evidenceIds":[],"sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","sourceLabel":"Google DeepMind · Gemini 3.1 Pro model card"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"google-gemini31","sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","observedAt":"2026-02-19","publishedAt":"2026-02-19","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-gemini-3-1-pro-swebench-verified","modelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","provider":"Google","familyId":"google/gemini-3.1","endpointId":null,"harnessId":"unspecified-reported","harnessName":"Provider-reported (harness unspecified)","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"swebench-verified@verified","metricId":"resolved","metric":"Resolved","value":80.6,"rawValue":"80.6%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"single attempt · model card"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"system","evidence":{"level":"B","sourceIds":["google-gemini31"],"evidenceIds":[],"sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","sourceLabel":"Google DeepMind · Gemini 3.1 Pro model card"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"google-gemini31","sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","observedAt":"2026-02-19","publishedAt":"2026-02-19","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-gemini-3-1-pro-mmmu","modelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","provider":"Google","familyId":"google/gemini-3.1","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"mmmu","benchmarkName":"MMMU","benchmarkVersion":"mmmu@v1","metricId":"accuracy","metric":"准确率","value":80.5,"rawValue":"80.5%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU-Pro · no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["google-gemini31"],"evidenceIds":[],"sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","sourceLabel":"Google DeepMind · Gemini 3.1 Pro model card"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"google-gemini31","sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","observedAt":"2026-02-19","publishedAt":"2026-02-19","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-gemini-3-1-pro-hle","modelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","provider":"Google","familyId":"google/gemini-3.1","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"hle@v1","metricId":"accuracy","metric":"准确率","value":44.4,"rawValue":"44.4%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"full set · no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["google-gemini31"],"evidenceIds":[],"sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","sourceLabel":"Google DeepMind · Gemini 3.1 Pro model card"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"google-gemini31","sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","observedAt":"2026-02-19","publishedAt":"2026-02-19","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-gemini-3-1-pro-gpqa-diamond","modelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","provider":"Google","familyId":"google/gemini-3.1","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"gpqa-diamond@v1","metricId":"accuracy","metric":"准确率","value":94.3,"rawValue":"94.3%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools · thinking high"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["google-gemini31"],"evidenceIds":[],"sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","sourceLabel":"Google DeepMind · Gemini 3.1 Pro model card"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"google-gemini31","sourceUrl":"https://deepmind.google/models/model-cards/gemini-3-1-pro/","observedAt":"2026-02-19","publishedAt":"2026-02-19","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-qwen3-max-thinking-swebench-verified","modelId":"qwen/qwen3-max@thinking","modelName":"Qwen3-Max-Thinking","provider":"Qwen","familyId":"qwen/qwen3-max","endpointId":null,"harnessId":"unspecified-reported","harnessName":"Provider-reported (harness unspecified)","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"swebench-verified@verified","metricId":"resolved","metric":"Resolved","value":75.3,"rawValue":"75.3%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"official setting · no custom scaffold"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"system","evidence":{"level":"B","sourceIds":["qwen3-max"],"evidenceIds":[],"sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","observedAt":"2026-01-15","publishedAt":"2026-01-15","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-qwen3-max-thinking-mmlu-pro","modelId":"qwen/qwen3-max@thinking","modelName":"Qwen3-Max-Thinking","provider":"Qwen","familyId":"qwen/qwen3-max","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersion":"mmlu-pro@v1","metricId":"accuracy","metric":"准确率","value":85.7,"rawValue":"85.7%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"官方横向表"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["qwen3-max"],"evidenceIds":[],"sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","observedAt":"2026-01-15","publishedAt":"2026-01-15","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-qwen3-max-thinking-livecodebench","modelId":"qwen/qwen3-max@thinking","modelName":"Qwen3-Max-Thinking","provider":"Qwen","familyId":"qwen/qwen3-max","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"livecodebench","benchmarkName":"LiveCodeBench","benchmarkVersion":"livecodebench@v5","metricId":"pass@1","metric":"Pass@1","value":85.9,"rawValue":"85.9%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"LiveCodeBench v6"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["qwen3-max"],"evidenceIds":[],"sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","observedAt":"2026-01-15","publishedAt":"2026-01-15","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-qwen3-max-thinking-hle","modelId":"qwen/qwen3-max@thinking","modelName":"Qwen3-Max-Thinking","provider":"Qwen","familyId":"qwen/qwen3-max","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"hle@v1","metricId":"accuracy","metric":"准确率","value":30.2,"rawValue":"30.2%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"text subset · no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["qwen3-max"],"evidenceIds":[],"sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","observedAt":"2026-01-15","publishedAt":"2026-01-15","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-qwen3-max-thinking-gpqa-diamond","modelId":"qwen/qwen3-max@thinking","modelName":"Qwen3-Max-Thinking","provider":"Qwen","familyId":"qwen/qwen3-max","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"gpqa-diamond@v1","metricId":"accuracy","metric":"准确率","value":87.4,"rawValue":"87.4%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"官方横向表 · no custom scaffold"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["qwen3-max"],"evidenceIds":[],"sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","sourceLabel":"Qwen · Qwen3-Max-Thinking"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"qwen3-max","sourceUrl":"https://qwen.ai/blog?id=qwen3-max-thinking","observedAt":"2026-01-15","publishedAt":"2026-01-15","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-grok-4-fast-livecodebench","modelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","provider":"xAI","familyId":"xai/grok-4","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"livecodebench","benchmarkName":"LiveCodeBench","benchmarkVersion":"livecodebench@v5","metricId":"pass@1","metric":"Pass@1","value":80.0,"rawValue":"80.0%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"Jan–May rolling window"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["xai-grok4fast"],"evidenceIds":[],"sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","observedAt":"2025-09-19","publishedAt":"2025-09-19","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-grok-4-fast-hle","modelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","provider":"xAI","familyId":"xai/grok-4","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"hle@v1","metricId":"accuracy","metric":"准确率","value":20.0,"rawValue":"20.0%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["xai-grok4fast"],"evidenceIds":[],"sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","observedAt":"2025-09-19","publishedAt":"2025-09-19","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-grok-4-fast-gpqa-diamond","modelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","provider":"xAI","familyId":"xai/grok-4","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"gpqa-diamond@v1","metricId":"accuracy","metric":"准确率","value":85.7,"rawValue":"85.7%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["xai-grok4fast"],"evidenceIds":[],"sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","observedAt":"2025-09-19","publishedAt":"2025-09-19","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-grok-4-fast-aime-2025","modelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","provider":"xAI","familyId":"xai/grok-4","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"aime-2025","benchmarkName":"AIME 2025","benchmarkVersion":"aime-2025@v1","metricId":"pass@1","metric":"Pass@1","value":92.0,"rawValue":"92.0%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["xai-grok4fast"],"evidenceIds":[],"sourceUrl":"https://x.ai/news/grok-4-fast","sourceLabel":"xAI · Grok 4 Fast"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"xai-grok4fast","sourceUrl":"https://x.ai/news/grok-4-fast","observedAt":"2025-09-19","publishedAt":"2025-09-19","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-gpt-5-swebench-verified","modelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","provider":"OpenAI","familyId":"openai/gpt-5","endpointId":null,"harnessId":"unspecified-reported","harnessName":"Provider-reported (harness unspecified)","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"swebench-verified@verified","metricId":"resolved","metric":"Resolved","value":74.9,"rawValue":"74.9%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"pass@1 · n=477"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"system","evidence":{"level":"B","sourceIds":["openai-gpt5"],"evidenceIds":[],"sourceUrl":"https://openai.com/index/introducing-gpt-5/","sourceLabel":"OpenAI · Introducing GPT-5"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"openai-gpt5","sourceUrl":"https://openai.com/index/introducing-gpt-5/","observedAt":"2025-08-07","publishedAt":"2025-08-07","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"OpenAI 使用 477 个内部验证子集；官方榜单的 500 题结果不可直接替代。","legacyMigrated":true},{"id":"legacy-gpt-5-mmmu","modelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","provider":"OpenAI","familyId":"openai/gpt-5","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"mmmu","benchmarkName":"MMMU","benchmarkVersion":"mmmu@v1","metricId":"accuracy","metric":"准确率","value":84.2,"rawValue":"84.2%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU · no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["openai-gpt5"],"evidenceIds":[],"sourceUrl":"https://openai.com/index/introducing-gpt-5/","sourceLabel":"OpenAI · Introducing GPT-5"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"openai-gpt5","sourceUrl":"https://openai.com/index/introducing-gpt-5/","observedAt":"2025-08-07","publishedAt":"2025-08-07","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-gpt-5-gpqa-diamond","modelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","provider":"OpenAI","familyId":"openai/gpt-5","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"gpqa-diamond@v1","metricId":"accuracy","metric":"准确率","value":88.4,"rawValue":"88.4%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"GPT-5 pro · no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["openai-gpt5"],"evidenceIds":[],"sourceUrl":"https://openai.com/index/introducing-gpt-5/","sourceLabel":"OpenAI · Introducing GPT-5"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"openai-gpt5","sourceUrl":"https://openai.com/index/introducing-gpt-5/","observedAt":"2025-08-07","publishedAt":"2025-08-07","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"该值是 GPT-5 pro 变体，不应与普通 GPT-5 endpoint 混为同一 release。","legacyMigrated":true},{"id":"legacy-gpt-5-aime-2025","modelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","provider":"OpenAI","familyId":"openai/gpt-5","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"aime-2025","benchmarkName":"AIME 2025","benchmarkVersion":"aime-2025@v1","metricId":"pass@1","metric":"Pass@1","value":94.6,"rawValue":"94.6%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools · high"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["openai-gpt5"],"evidenceIds":[],"sourceUrl":"https://openai.com/index/introducing-gpt-5/","sourceLabel":"OpenAI · Introducing GPT-5"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"openai-gpt5","sourceUrl":"https://openai.com/index/introducing-gpt-5/","observedAt":"2025-08-07","publishedAt":"2025-08-07","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-gemini-2-5-pro-mmmu","modelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","provider":"Google","familyId":"google/gemini-2.5","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"mmmu","benchmarkName":"MMMU","benchmarkVersion":"mmmu@v1","metricId":"accuracy","metric":"准确率","value":81.7,"rawValue":"81.7%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU · single attempt"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["google-gemini25"],"evidenceIds":[],"sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","observedAt":"2025-06-05","publishedAt":"2025-06-05","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-gemini-2-5-pro-gpqa-diamond","modelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","provider":"Google","familyId":"google/gemini-2.5","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"gpqa-diamond@v1","metricId":"accuracy","metric":"准确率","value":86.4,"rawValue":"86.4%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"single attempt · no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["google-gemini25"],"evidenceIds":[],"sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","observedAt":"2025-06-05","publishedAt":"2025-06-05","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-gemini-2-5-pro-aime-2025","modelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","provider":"Google","familyId":"google/gemini-2.5","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"aime-2025","benchmarkName":"AIME 2025","benchmarkVersion":"aime-2025@v1","metricId":"pass@1","metric":"Pass@1","value":88.0,"rawValue":"88.0%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"single attempt · no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["google-gemini25"],"evidenceIds":[],"sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","observedAt":"2025-06-05","publishedAt":"2025-06-05","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-deepseek-r1-0528-mmlu-pro","modelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","provider":"DeepSeek","familyId":"deepseek/r1","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersion":"mmlu-pro@v1","metricId":"accuracy","metric":"准确率","value":81.2,"rawValue":"81.2%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"R1-0528 · reported"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["deepseek-r1-0528"],"evidenceIds":[],"sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API · R1-0528 update"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"deepseek-r1-0528","sourceUrl":"https://api-docs.deepseek.com/updates/","observedAt":"2025-05-28","publishedAt":"2025-05-28","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-deepseek-r1-0528-gpqa-diamond","modelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","provider":"DeepSeek","familyId":"deepseek/r1","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"gpqa-diamond@v1","metricId":"accuracy","metric":"准确率","value":68.4,"rawValue":"68.4%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"R1-0528 · reported"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["deepseek-r1-0528"],"evidenceIds":[],"sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API · R1-0528 update"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"deepseek-r1-0528","sourceUrl":"https://api-docs.deepseek.com/updates/","observedAt":"2025-05-28","publishedAt":"2025-05-28","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-deepseek-r1-0528-aime-2025","modelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","provider":"DeepSeek","familyId":"deepseek/r1","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"aime-2025","benchmarkName":"AIME 2025","benchmarkVersion":"aime-2025@v1","metricId":"pass@1","metric":"Pass@1","value":87.5,"rawValue":"87.5%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"R1-0528 · reported"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["deepseek-r1-0528"],"evidenceIds":[],"sourceUrl":"https://api-docs.deepseek.com/updates/","sourceLabel":"DeepSeek API · R1-0528 update"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"deepseek-r1-0528","sourceUrl":"https://api-docs.deepseek.com/updates/","observedAt":"2025-05-28","publishedAt":"2025-05-28","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-claude-opus-4-terminal-bench","modelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","provider":"Anthropic","familyId":"anthropic/claude-opus","endpointId":null,"harnessId":"unspecified-reported","harnessName":"Provider-reported (harness unspecified)","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersion":"terminal-bench@2.0","metricId":"pass_rate","metric":"Pass rate","value":43.2,"rawValue":"43.2%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"Claude Code scaffold"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"system","evidence":{"level":"B","sourceIds":["anthropic-claude4"],"evidenceIds":[],"sourceUrl":"https://www.anthropic.com/news/claude-4","sourceLabel":"Anthropic · Introducing Claude 4"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"anthropic-claude4","sourceUrl":"https://www.anthropic.com/news/claude-4","observedAt":"2025-05-22","publishedAt":"2025-05-22","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-claude-opus-4-swebench-verified","modelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","provider":"Anthropic","familyId":"anthropic/claude-opus","endpointId":null,"harnessId":"unspecified-reported","harnessName":"Provider-reported (harness unspecified)","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"swebench-verified@verified","metricId":"resolved","metric":"Resolved","value":72.5,"rawValue":"72.5%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"bash/editor scaffold · pass@1"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"system","evidence":{"level":"B","sourceIds":["anthropic-claude4"],"evidenceIds":[],"sourceUrl":"https://www.anthropic.com/news/claude-4","sourceLabel":"Anthropic · Introducing Claude 4"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"anthropic-claude4","sourceUrl":"https://www.anthropic.com/news/claude-4","observedAt":"2025-05-22","publishedAt":"2025-05-22","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-claude-opus-4-mmmu","modelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","provider":"Anthropic","familyId":"anthropic/claude-opus","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"mmmu","benchmarkName":"MMMU","benchmarkVersion":"mmmu@v1","metricId":"accuracy","metric":"准确率","value":76.5,"rawValue":"76.5%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"MMMU validation · extended thinking"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["anthropic-claude4"],"evidenceIds":[],"sourceUrl":"https://www.anthropic.com/news/claude-4","sourceLabel":"Anthropic · Introducing Claude 4"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"anthropic-claude4","sourceUrl":"https://www.anthropic.com/news/claude-4","observedAt":"2025-05-22","publishedAt":"2025-05-22","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-claude-opus-4-gpqa-diamond","modelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","provider":"Anthropic","familyId":"anthropic/claude-opus","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"gpqa-diamond@v1","metricId":"accuracy","metric":"准确率","value":83.3,"rawValue":"83.3%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"extended thinking · 64K"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["anthropic-claude4"],"evidenceIds":[],"sourceUrl":"https://www.anthropic.com/news/claude-4","sourceLabel":"Anthropic · Introducing Claude 4"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"anthropic-claude4","sourceUrl":"https://www.anthropic.com/news/claude-4","observedAt":"2025-05-22","publishedAt":"2025-05-22","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-claude-opus-4-aime-2025","modelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","provider":"Anthropic","familyId":"anthropic/claude-opus","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"aime-2025","benchmarkName":"AIME 2025","benchmarkVersion":"aime-2025@v1","metricId":"pass@1","metric":"Pass@1","value":75.5,"rawValue":"75.5%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"single attempt · extended thinking"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["anthropic-claude4"],"evidenceIds":[],"sourceUrl":"https://www.anthropic.com/news/claude-4","sourceLabel":"Anthropic · Introducing Claude 4"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"anthropic-claude4","sourceUrl":"https://www.anthropic.com/news/claude-4","observedAt":"2025-05-22","publishedAt":"2025-05-22","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"发布表另列出并行 test-time compute 的 90.0%，此处保留单次值。","legacyMigrated":true},{"id":"legacy-gemini-2-5-pro-swebench-verified","modelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","provider":"Google","familyId":"google/gemini-2.5","endpointId":null,"harnessId":"unspecified-reported","harnessName":"Provider-reported (harness unspecified)","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"swebench-verified@verified","metricId":"resolved","metric":"Resolved","value":63.8,"rawValue":"63.8%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"custom agent setup"},"subject":{"type":"system","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"system","evidence":{"level":"B","sourceIds":["google-gemini25"],"evidenceIds":[],"sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","observedAt":"2025-03-25","publishedAt":"2025-03-25","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-gemini-2-5-pro-livecodebench","modelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","provider":"Google","familyId":"google/gemini-2.5","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"livecodebench","benchmarkName":"LiveCodeBench","benchmarkVersion":"livecodebench@v5","metricId":"pass@1","metric":"Pass@1","value":70.4,"rawValue":"70.4%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"V5 · rolling window"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["google-gemini25"],"evidenceIds":[],"sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","observedAt":"2025-03-25","publishedAt":"2025-03-25","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true},{"id":"legacy-gemini-2-5-pro-hle","modelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","provider":"Google","familyId":"google/gemini-2.5","endpointId":null,"harnessId":"model-only","harnessName":"Model-only / single turn","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"hle@v1","metricId":"accuracy","metric":"准确率","value":18.8,"rawValue":"18.8%","unit":"%","protocol":{"shots":null,"temperature":null,"top_p":null,"tools":null,"reasoning_mode":null,"raw_setting":"no tools"},"subject":{"type":"model","system_id":null,"agent_id":null,"scaffold":null},"subjectType":"model","evidence":{"level":"B","sourceIds":["google-gemini25"],"evidenceIds":[],"sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","sourceLabel":"Google · Gemini 2.5 model card/blog"},"evidenceLevel":"B","evidenceIds":[],"comparability":"conditional","status":"reported","sourceId":"google-gemini25","sourceUrl":"https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/","observedAt":"2025-03-25","publishedAt":"2025-03-25","uncertainty":null,"sampleSize":null,"qualityFlags":[],"preferred":true,"notes":"Migrated from legacy nested seed; protocol detail is limited.","legacyMigrated":true}],"presets":[{"id":"frontier-current","label":"Frontier / current","description":"当前旗舰与重要 frontier release；各 benchmark 独立显示，不计算总分。","default":true,"model_filter":{"status":["active","preview","restricted"]},"benchmark_ids":["gpqa-diamond","mmlu-pro","hle","swebench-pro","terminal-bench","mmmu-pro"],"sort":"release_desc"},{"id":"latest-vs-previous","label":"Latest vs previous","description":"同一 family 的最新 release 与上一代；保留历史而不让旧模型占默认榜。","model_filter":{"include_tags":["旗舰","上一代"]},"benchmark_ids":["gpqa-diamond","aime-2026","aime-2025","mmlu-pro","livecodebench","swebench-verified"],"sort":"family_release"},{"id":"flash-fast","label":"Flash / fast","description":"按 Flash、mini、highspeed 或 lower-latency variant 筛选；同时展示质量、成本和延迟。","model_filter":{"include_tags":["Flash","高速","小模型","效率"]},"benchmark_ids":["gpqa-diamond","livecodebench","terminal-bench","bfcl"],"sort":"efficiency_pareto","dimensions":["score","input_cost","output_cost","ttft","output_tokens_per_second"]},{"id":"small-efficient","label":"Small / efficient","description":"优先按 active parameters、显存、吞吐和单位任务成本比较；未知参数不猜。","model_filter":{"include_tags":["小模型","效率","高速"]},"benchmark_ids":["mmlu-pro","gpqa-diamond","livecodebench","bfcl"],"sort":"active_params"},{"id":"math-science","label":"Math / science","description":"数学与研究生科学推理；保留 pass@k、工具和采样设置。","benchmark_ids":["aime-2026","aime-2025","hmmt","frontiermath","gpqa-diamond","hle"],"model_filter":{"include_tags":["推理","数学"]},"sort":"benchmark_column"},{"id":"static-coding","label":"Static coding","description":"无 agent loop 的代码能力；固定题目窗口与解码协议。","benchmark_ids":["livecodebench","mmlu-pro"],"harness_ids":["model-only","lm-eval"],"sort":"benchmark_column"},{"id":"coding-agent","label":"Coding Agent","description":"固定 harness 后比较端到端软件工程结果。","mode":"runs","benchmark_ids":["swebench-verified","swebench-pro","deepswe","nl2repo","terminal-bench"],"harness_ids":["swebench-official","mini-swe-agent","terminus-2","claude-code","codex","kimi-code","deepseek-harness"],"sort":"benchmark_column"},{"id":"harness-ab","label":"Same model × harness","description":"固定 model，切换 agent/harness，观察 scaffold effect；不与 model-only 分数混排。","mode":"runs","benchmark_ids":["swebench-verified","terminal-bench","deepswe"],"compare":{"axis":"harness","group_by":"model_id"},"sort":"harness"},{"id":"tool-workflow","label":"Tools / workflow","description":"函数调用、业务流程和长程工具使用；同时看成功率与 pass^k。","mode":"runs","benchmark_ids":["bfcl","tau-bench","toolathlon","gdpval-aa"],"sort":"benchmark_column"},{"id":"multimodal-computer","label":"Multimodal / computer use","description":"图像、视频和桌面环境；输入模态与 GUI scaffold 必须明确。","mode":"runs","benchmark_ids":["mmmu","mmmu-pro","mathvista","video-mme","osworld"],"model_filter":{"modalities_any":["image","video","audio"]},"sort":"benchmark_column"},{"id":"long-context","label":"Long context","description":"按 context bucket 查看检索/推理曲线，不用一个最大窗口数字替代实际表现。","benchmark_ids":["ruler","longbench"],"model_filter":{"min_context_window":128000},"sort":"context_window"},{"id":"open-weight-deployment","label":"Open weight / deployment","description":"开放权重模型的质量、license、total/active parameters、量化和部署成本。","model_filter":{"open_weights":true},"benchmark_ids":["gpqa-diamond","mmlu-pro","livecodebench","swebench-verified"],"sort":"active_params","dimensions":["score","params_total","params_active","license","vram","throughput"]},{"id":"chinese-multilingual","label":"Chinese / multilingual","description":"中文与多语言单独比较，避免英语总分掩盖语言差异。","benchmark_ids":["ceval","cmmlu","superclue","mmmlu"],"model_filter":{"modalities_any":["text"]},"sort":"benchmark_column"},{"id":"reliability-safety","label":"Reliability / safety","description":"重复运行、拒答、事实性和网络安全；优先展示置信区间与失败类型。","benchmark_ids":["simpleqa","tau-bench","cybergym","hle"],"sort":"benchmark_column","dimensions":["score","pass_k","pass_power_k","uncertainty","failure_modes"]},{"id":"scale-watchlist","label":"Scale watchlist","description":"只收 source-backed 参数；10T 目前是 watchlist，不对未知规模做排名。","model_filter":{"min_params_total":1000000000000},"benchmark_ids":["gpqa-diamond","frontiermath","swebench-pro","terminal-bench"],"sort":"params_total","note":"total parameters and active parameters are separate fields; unknown is not zero"},{"id":"arena-preference","label":"Arena / human preference","description":"Arena 人类偏好 rating 单独展示；按 arena category、snapshot date、votes 与 CI 比较，不与 accuracy 混合。","benchmark_ids":["arena-text","arena-vision","arena-document"],"sort":"benchmark_column","dimensions":["rating","rating_ci","vote_count","category","snapshot_date"]},{"id":"arena-agent","label":"Arena / agent workflows","description":"Agent Arena IPS 与 WebDev/Search system runs；固定 harness/context 后比较，不能当作裸模型分数。","mode":"runs","benchmark_ids":["arena-agent","arena-webdev","arena-search"],"harness_ids":["arena-agent","arena-human-preference"],"sort":"benchmark_column","dimensions":["ips","rating","confidence_interval","observation_count","session_count"]},{"id":"ale-workflows","label":"ALE / professional agent workflows","description":"Agents' Last Exam ALE-V1；按 split、source harness、effort、pass rate、partial score、成本、运行时间和 token 细节比较，不将 system run 当作裸模型分数。","mode":"runs","benchmark_ids":["agents-last-exam"],"harness_ids":["agents-last-exam","codex","claude-code","kimi-code"],"sort":"benchmark_column","dimensions":["pass_rate","avg_score","split","harness","harness_variant","runs","tasks","passes","cost","runtime","tokens"]},{"id":"algorithm-engineering","label":"Algorithm engineering","description":"ALE-Bench 的长程、连续评分算法工程；独立于 Agents' Last Exam 的专业工作流 pass rate。","benchmark_ids":["ale-bench"],"sort":"benchmark_column","dimensions":["performance","rank","self_refine","cost","total_tokens"]}],"stats":{"catalogModels":100,"scoredModels":9,"catalogOnlyModels":91,"benchmarks":53,"featuredBenchmarks":12,"sources":110,"harnesses":15,"presets":19,"runs":47,"systemRuns":18,"observations":47,"numericObservations":47,"observedCells":29,"coverage":0.0055,"coveragePct":0.5,"legacyMigrated":32,"modelProfiles":100,"modelProfilesMissing":0},"publicEvidence":[{"id":"pub-ff43ab3710001fb63ba2cbba","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":7.978474570120461,"rawValue":0.07978474570120461,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[578];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ff2f5ab1861f9ab70c83e8e0","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.064451042627894,"rawValue":0.5206445104262789,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[663];split=unlicensed/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-f1fdeeb18f493765ec4a03b6","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":15.451296049520938,"rawValue":0.15451296049520938,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[323];split=licensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-f185b5bd79c17d6d982d58b5","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":67.36742823508564,"rawValue":0.6736742823508564,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[193];split=full/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-e5b9e54d80a53ac4621a920c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.064451042627894,"rawValue":0.5206445104262789,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[427];split=unlicensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-e002fa0b7775cda6b162ba3b","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.721348774274084,"rawValue":0.4372134877427408,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[67];split=full/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-de19737888e5d736334e5a13","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.7344624718011,"rawValue":0.507344624718011,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[362];split=linux_only;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-dddb41967a7b16ff8d81ef36","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":53.31379784041669,"rawValue":0.5331379784041669,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[423];split=unlicensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-d9b46f9679c8d135f42b4a79","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.056329147969095,"rawValue":0.40056329147969094,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[691];split=unlicensed/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-d438de9ba7bcee6b5368d978","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":49.41481482602146,"rawValue":0.49414814826021464,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[246];split=full/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-d2de271165f71939c3e85403","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":75.62644213793284,"rawValue":0.7562644213793285,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[180];split=full/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-ce887a04b39625cb59a09ee9","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":11.7529387058429,"rawValue":0.11752938705842898,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[559];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-ca34cd5bfb5f1e0052606a68","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":49.41481482602146,"rawValue":0.49414814826021464,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[10];split=full;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-c40c2b5ded82b4717e0f1024","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.993413212460641,"rawValue":0.10993413212460641,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[143];split=full/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-c38ce812ed78e9c3c87fe010","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.056329147969095,"rawValue":0.40056329147969094,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[455];split=unlicensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-bd39df45b225832ffe65397e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.53164625672212,"rawValue":0.4253164625672212,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[71];split=full/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-bd2929b576735cefd268d960","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.37352082410918,"rawValue":0.5037352082410917,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[9];split=full;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-bbeda83374aef769ba0264ab","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.12876688214367,"rawValue":0.46128766882143674,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[16];split=full;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-b444f56c8cfaff3a5528bf15","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.61045019913022,"rawValue":0.3361045019913022,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[401];split=linux_only;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-ae8990cc48b17bb9bb8e32df","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.12876688214367,"rawValue":0.46128766882143674,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[252];split=full/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-ac68eeee2b918ed7dddb14a4","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.77055625187485,"rawValue":0.4177055625187485,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[490];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-abca6778d092f9e791cbe141","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.435207855503172,"rawValue":0.2343520785550317,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[304];split=licensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-aa3cb750ae69a1a34c7f42f9","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":32.80611890351934,"rawValue":0.3280611890351934,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[700];split=unlicensed/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-a810fd86eb7e27f4779ab497","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":29.636355630497178,"rawValue":0.2963635563049718,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[517];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-a5de15466786bf2f839f4c8e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":9.622840053229611,"rawValue":0.09622840053229612,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[568];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":25},{"id":"pub-a387939fd75e48a98193c9fc","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.970741091203196,"rawValue":0.10970741091203197,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[144];split=full/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":26},{"id":"pub-99ca04c05184c4be617585b9","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":30.073895909931007,"rawValue":0.3007389590993101,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[97];split=full/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":27},{"id":"pub-96f18889d45a18886f363ed0","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":15.051983271916239,"rawValue":0.1505198327191624,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[547];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":28},{"id":"pub-961a568bf379e956fd97f8a3","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":47.899186380675914,"rawValue":0.4789918638067592,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[434];split=unlicensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":29},{"id":"pub-91ab6bf0373a9d75a8791d22","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.273103955698325,"rawValue":0.48273103955698327,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[222];split=full/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":30},{"id":"pub-8cf99679f407338ab7156c5f","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":31.023834901975544,"rawValue":0.31023834901975544,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[282];split=full/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":31},{"id":"pub-860f794638531334ebb2f962","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":75.62644213793284,"rawValue":0.7562644213793285,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[598];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":32},{"id":"pub-8212a4cb1e34bd341ad4ad50","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":31.023834901975544,"rawValue":0.31023834901975544,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[46];split=full;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":33},{"id":"pub-7867c2ff650149d8c18635c7","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":24.564591975351032,"rawValue":0.24564591975351033,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[524];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":34},{"id":"pub-6ec386e220ad565db5f7c340","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.37352082410918,"rawValue":0.5037352082410917,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[245];split=full/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":35},{"id":"pub-6e77139565a32ea7a46a3043","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":67.36742823508564,"rawValue":0.6736742823508564,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[611];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":36},{"id":"pub-6d1e6f3cc5f545001288da0b","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":74.3547028673177,"rawValue":0.743547028673177,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[182];split=full/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":37},{"id":"pub-64415c51f00e0f41e1f20621","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":32.80611890351934,"rawValue":0.3280611890351934,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[464];split=unlicensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":38},{"id":"pub-632d0d6ba52895a590c55f8a","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.50041500197,"rawValue":0.3850041500197,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[272];split=full/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":39},{"id":"pub-5e0760df54c3e20eae869b06","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":18.55642458507263,"rawValue":0.1855642458507263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[319];split=licensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":40},{"id":"pub-5d6342079c87b1c6e2eaf27c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":7.960916226247187,"rawValue":0.07960916226247187,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[579];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":41},{"id":"pub-5be0b289b950bd0740611d62","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.395747737317528,"rawValue":0.10395747737317527,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[149];split=full/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":42},{"id":"pub-5b5f954bc6def8301500371c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.712583069964325,"rawValue":0.41712583069964326,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[73];split=full/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":43},{"id":"pub-577ef54bde84ac7ad2d8272a","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":60.92947556539312,"rawValue":0.6092947556539312,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[207];split=full/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":44},{"id":"pub-56f53630077d22f8d7c4bb72","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":24.891449790392855,"rawValue":0.24891449790392856,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[103];split=full/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":45},{"id":"pub-531eeee3df6d32c488e347c5","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":53.31379784041669,"rawValue":0.5331379784041669,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[659];split=unlicensed/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":46},{"id":"pub-52e39d44dc1451073f76396f","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":47.755529230158054,"rawValue":0.47755529230158056,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[370];split=linux_only;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":47},{"id":"pub-4db23db1c55dd48d8f697b71","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.684515433258309,"rawValue":0.1268451543325831,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[327];split=licensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":48},{"id":"pub-480f179605d41209a03a67d4","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.56378222921257,"rawValue":0.4356378222921257,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[484];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":49},{"id":"pub-45a0d469978fec9bb0616caf","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.273103955698325,"rawValue":0.48273103955698327,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[640];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":50},{"id":"pub-3c8a1944ced12ef65b2c38a0","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.38739062483173,"rawValue":0.5238739062483173,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[357];split=linux_only;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":51},{"id":"pub-388bd27514db2d168f4538ce","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":60.92947556539312,"rawValue":0.6092947556539312,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[625];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":52},{"id":"pub-35591a8371b1502f86442299","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.50041500197,"rawValue":0.3850041500197,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[36];split=full;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":53},{"id":"pub-35444cf18c62af42d87a1d26","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.18575126439045,"rawValue":0.41185751264390447,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[389];split=linux_only;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":54},{"id":"pub-3365a605bff0b0db17c5902b","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.890663199918114,"rawValue":0.08890663199918114,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[154];split=full/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":55},{"id":"pub-32bf3da9fbc76d974efba0e4","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.178194518550498,"rawValue":0.08178194518550498,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[334];split=licensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":56},{"id":"pub-30a681b01fe903c0a575e94f","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.33841810461951,"rawValue":0.4433841810461951,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[482];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":57},{"id":"pub-2116bd74dc3c209356b1edfc","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":47.899186380675914,"rawValue":0.4789918638067592,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[670];split=unlicensed/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":58},{"id":"pub-1308eefacfc0293fb0354bff","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":74.3547028673177,"rawValue":0.743547028673177,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[600];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":59},{"id":"pub-00d79827380c1e42a28d8440","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.96761918512072,"rawValue":0.1496761918512072,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[129];split=full/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":60},{"id":"pub-fe28594badc57e44f221fe9f","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":35.82089552238806,"rawValue":0.3582089552238806,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[193];split=full/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fa8035be3a321e36b36250a7","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":43.28358208955223,"rawValue":0.43283582089552236,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[182];split=full/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-f758ca08b30d50b57c728a4c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":31.20567375886525,"rawValue":0.3120567375886525,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[659];split=unlicensed/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-f68fd72068c7b893e3751ec8","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.57142857142857,"rawValue":0.2857142857142857,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[357];split=linux_only;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-f4a51ce97d490c8d62e7cd72","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.76595744680851,"rawValue":0.1276595744680851,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[464];split=unlicensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-f41d071e980ef4c2e045f4ed","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.09090909090909,"rawValue":0.2909090909090909,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[71];split=full/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-effc46595853b26de7618c2d","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.0,"rawValue":0.12,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[517];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-ed73792a27ab9d114e0535d4","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.842105263157894,"rawValue":0.11842105263157894,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[282];split=full/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-e8ab76b9fa7b0bb0b2125b5e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.388059701492537,"rawValue":0.22388059701492538,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[640];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-dfc9b9af3653b0073130ecac","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.095238095238095,"rawValue":0.18095238095238095,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[389];split=linux_only;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-db04de1c6450c0db49dffa3f","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[578];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-d8719fb0c243e2f53834302d","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.342105263157894,"rawValue":0.24342105263157895,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[252];split=full/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-d69e558d871c60b09c7d1743","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[154];split=full/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-d19648c59c7731f016f60ebe","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":7.2727272727272725,"rawValue":0.07272727272727272,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[103];split=full/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-c5429a3fb99a4437e375a73d","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[319];split=licensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-c402160d49e5a711ffee920b","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[579];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-c3ad68c1b5bb805c071a9167","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":31.20567375886525,"rawValue":0.3120567375886525,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[423];split=unlicensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-bfdf29433b8d8cda37ea40a8","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.761904761904763,"rawValue":0.24761904761904763,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[370];split=linux_only;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-be9c83647b5cadc91c1f5a51","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[144];split=full/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-bb4a1928a152a58ea1519b25","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.76595744680851,"rawValue":0.1276595744680851,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[700];split=unlicensed/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-adc6d0b2c403d338dbba852e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.27272727272727,"rawValue":0.2727272727272727,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[67];split=full/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-a803ddac5a96ea86071d03fa","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.24113475177305,"rawValue":0.2624113475177305,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[434];split=unlicensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-a6108154d2cc500d9a048903","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.62411347517731,"rawValue":0.3262411347517731,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[663];split=unlicensed/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-a5ec43e616951c2c22ba389f","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.342105263157894,"rawValue":0.24342105263157895,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[16];split=full;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-9f6f4ecd7d4744a1c5424e11","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.000000000000004,"rawValue":0.28,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[484];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":25},{"id":"pub-9e8efe3e531a8187bcc4bd84","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.523809523809526,"rawValue":0.29523809523809524,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[362];split=linux_only;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":26},{"id":"pub-97518880af13d2cb8450a090","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.842105263157894,"rawValue":0.11842105263157894,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[46];split=full;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":27},{"id":"pub-8ec819d8e58851167c118606","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.24113475177305,"rawValue":0.2624113475177305,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[670];split=unlicensed/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":28},{"id":"pub-8dc731de6e3aae718d4ce351","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[143];split=full/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":29},{"id":"pub-8a525852e7bdedf37952a121","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[129];split=full/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":30},{"id":"pub-802963b4e4f9f455b401ee8e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":49.25373134328358,"rawValue":0.4925373134328358,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[180];split=full/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":31},{"id":"pub-7bca23ca7529fa12cd13483f","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.947368421052634,"rawValue":0.2894736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[245];split=full/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":32},{"id":"pub-781978574287f8e6a11065c8","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":30.0,"rawValue":0.3,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[482];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":33},{"id":"pub-7818293b108b6a3f1ae60ec8","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.105263157894736,"rawValue":0.17105263157894737,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[272];split=full/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":34},{"id":"pub-77b454699f4866ffe99e2df2","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.62411347517731,"rawValue":0.3262411347517731,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[427];split=unlicensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":35},{"id":"pub-76b7cc72c5f3b3a7ddcfd699","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.850746268656714,"rawValue":0.29850746268656714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[625];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":36},{"id":"pub-75704703d97d1e20d637f22f","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[547];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":37},{"id":"pub-6a13bfa6c82952a17bb96cff","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":49.25373134328358,"rawValue":0.4925373134328358,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[598];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":38},{"id":"pub-6639dd12917a210eec8dc8d5","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[334];split=licensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":39},{"id":"pub-65d87e1330caf89dffc9e1a5","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.439716312056735,"rawValue":0.18439716312056736,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[455];split=unlicensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":40},{"id":"pub-5b415e98485b0d1afe527329","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.439716312056735,"rawValue":0.18439716312056736,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[691];split=unlicensed/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":41},{"id":"pub-5878e5fd66939f93605c0e11","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.27272727272727,"rawValue":0.2727272727272727,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[73];split=full/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":42},{"id":"pub-55bed99f5d7aac3dc4fb4cc3","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[568];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":43},{"id":"pub-5564be1af8e0056952b83d44","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.105263157894736,"rawValue":0.17105263157894737,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[36];split=full;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":44},{"id":"pub-54d0f88e53ec1d3c018b9a29","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":43.28358208955223,"rawValue":0.43283582089552236,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[600];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":45},{"id":"pub-46dedfe2be543e0bb80d27d3","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[149];split=full/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":46},{"id":"pub-4696e27490117893a78b959a","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.388059701492537,"rawValue":0.22388059701492538,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[222];split=full/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":47},{"id":"pub-41758eee66be105ffea44c0a","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":6.0,"rawValue":0.06,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[524];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":48},{"id":"pub-39b089a37a484bd595221d03","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":30.263157894736842,"rawValue":0.3026315789473684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[246];split=full/overall;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":49},{"id":"pub-25f804276ca8334d31523ead","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[327];split=licensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":50},{"id":"pub-1fe94478d5eff159732816ec","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[323];split=licensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":51},{"id":"pub-1ca4f98b812e597d1df1ef52","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":35.82089552238806,"rawValue":0.3582089552238806,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[611];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":52},{"id":"pub-1a3df7abcb8e6ccadb683560","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.727272727272727,"rawValue":0.12727272727272726,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[97];split=full/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":53},{"id":"pub-0eb2c8521d7b59bf6f7ee743","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[304];split=licensed;harness=codex;model=GPT-5.6-Luna;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":54},{"id":"pub-0d55cbe815c62c6801933073","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[559];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":55},{"id":"pub-0c104dc565e5cfb8dd74268a","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.947368421052634,"rawValue":0.2894736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[9];split=full;harness=codex;model=GPT-5.6-Luna;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":56},{"id":"pub-03a2db6248b975db950d9292","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.428571428571429,"rawValue":0.11428571428571428,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[401];split=linux_only;harness=codex;model=GPT-5.6-Luna;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":57},{"id":"pub-009ce552f5950e7d62d3d6b2","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":30.263157894736842,"rawValue":0.3026315789473684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[10];split=full;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":58},{"id":"pub-0095275f439366ceca18a45f","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.000000000000004,"rawValue":0.28,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[490];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Luna;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":59},{"id":"pub-0078961745ba978a118d4141","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT-5.6-Luna","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.850746268656714,"rawValue":0.29850746268656714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[207];split=full/near-term;harness=codex;model=GPT-5.6-Luna;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":60},{"id":"pub-fded58569e7e478e7cd940c9","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.61806657735013,"rawValue":0.1261806657735013,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[137];split=full/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fadea1877e295ac21ba9cfeb","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":18.922557053100295,"rawValue":0.18922557053100295,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[543];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-fac7ddb6d5db33448ed711cc","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":53.63939737297058,"rawValue":0.5363939737297058,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[354];split=linux_only;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-fa7d64ef7743bbe96513439a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":78.29241382323966,"rawValue":0.7829241382323966,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[178];split=full/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-f85f98c24b26a6fdc5418b43","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":53.137486902903696,"rawValue":0.531374869029037,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[355];split=linux_only;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-ef1aca12a05d260dfb61c72b","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":77.5482331051408,"rawValue":0.775482331051408,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[597];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-e94bd87d0a79e5b66d6fe09a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":55.954839666139776,"rawValue":0.5595483966613978,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[418];split=unlicensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-d90f6237b6e09c0b7ec130fb","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.81048909575152,"rawValue":0.4681048909575152,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[437];split=unlicensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-d442fd1e64efb1bb744ec407","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.10026942935828,"rawValue":0.5210026942935828,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[4];split=full;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-cd796305a7e11eae03bfa53c","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":54.91172987872457,"rawValue":0.5491172987872457,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[656];split=unlicensed/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-cc0e2f39ccb2515302852530","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":54.30423343738866,"rawValue":0.5430423343738866,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[422];split=unlicensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-cbbbaba2d18cffb992ba8d55","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":66.77679137696316,"rawValue":0.6677679137696316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[197];split=full/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-c81b61bafcc67abc06836d3a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":54.30423343738866,"rawValue":0.5430423343738866,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[658];split=unlicensed/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-c7776423a3a9b2f319170d8f","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.37286731699879,"rawValue":0.523728673169988,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[239];split=full/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-c3761fefb0ef48e0b5f90583","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.762780045204536,"rawValue":0.5276278004520454,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[237];split=full/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-c22ceb4c6b663d1e767d2cd0","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":53.61628463309977,"rawValue":0.5361628463309976,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[0];split=full;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-bda5768f50630b62e57a516a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.51610959401981,"rawValue":0.4451610959401981,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[66];split=full/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-bb8ee68f24773ba0950c8432","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.363827043507804,"rawValue":0.433638270435078,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[486];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-b734c120e9046d4447ea66a9","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.076492497958796,"rawValue":0.44076492497958797,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[483];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-b35d633e987d2bafa3c175c6","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.762780045204536,"rawValue":0.5276278004520454,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[1];split=full;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-b355b107d2b4368cde792f56","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":78.82473228071811,"rawValue":0.7882473228071811,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[177];split=full/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-b323823a7077827d730553d9","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":66.77679137696316,"rawValue":0.6677679137696316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[615];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-af6c8eb5f50556c16dddf401","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":75.61914737576828,"rawValue":0.7561914737576828,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[599];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-ae9c99a6259707a2844d70d0","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.315914835731926,"rawValue":0.12315914835731925,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[557];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-ae4d716a3ef4e3328980c416","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":51.215569563350606,"rawValue":0.512155695633506,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[359];split=linux_only;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":25},{"id":"pub-ac6522c275562d2fab85c165","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.014328647763186,"rawValue":0.4301432864776319,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[69];split=full/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":26},{"id":"pub-a3bd64d04637eef39ae007a5","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.849458053696015,"rawValue":0.23849458053696015,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[302];split=licensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":27},{"id":"pub-9e13cf05da4f2a586bfc3b7a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":16.44048385293532,"rawValue":0.1644048385293532,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[127];split=full/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":28},{"id":"pub-97a1e8589d01e11ee1e2da1d","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":78.82473228071811,"rawValue":0.7882473228071811,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[595];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":29},{"id":"pub-94e233dd8465c75cff2eaa4d","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.14034337333402,"rawValue":0.23140343373334019,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[305];split=licensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":30},{"id":"pub-9463536824de9af79701af65","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":55.954839666139776,"rawValue":0.5595483966613978,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[654];split=unlicensed/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":31},{"id":"pub-8d3e9d3d0215b8c91914cedb","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":19.65229899475815,"rawValue":0.19652298994758152,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[542];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":32},{"id":"pub-8981d38a6333094148e0616e","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.640261027768616,"rawValue":0.23640261027768616,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[303];split=licensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":33},{"id":"pub-812a879af8c83707e1805a6c","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":18.68668857038979,"rawValue":0.1868668857038979,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[125];split=full/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":34},{"id":"pub-7c91e22163e3fb1691d74292","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":75.61914737576828,"rawValue":0.7561914737576828,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[181];split=full/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":35},{"id":"pub-78a64a0e1845259556f3f94b","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":78.29241382323966,"rawValue":0.7829241382323966,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[596];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":36},{"id":"pub-774b8aeffef1a3b574546415","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":20.737264567839304,"rawValue":0.20737264567839306,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[539];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":37},{"id":"pub-730ccdcab5ef217788be3d4b","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.10026942935828,"rawValue":0.5210026942935828,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[240];split=full/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":38},{"id":"pub-71c822a953d9555581893fa8","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.285408160616505,"rawValue":0.39285408160616503,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[498];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":39},{"id":"pub-6aa11bb93c32f07192589271","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.09751802373446,"rawValue":0.4509751802373446,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[19];split=full;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":40},{"id":"pub-64435483a543f951c96fd396","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.37286731699879,"rawValue":0.523728673169988,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[3];split=full;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":41},{"id":"pub-60a46a4b17e13217cdeb0036","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":54.736328337091045,"rawValue":0.5473632833709104,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[421];split=unlicensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":42},{"id":"pub-588f62a4e071a4f774dcae1b","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":53.61628463309977,"rawValue":0.5361628463309976,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[236];split=full/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":43},{"id":"pub-51641bccfc664ba3e15ab90d","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.948989785767765,"rawValue":0.5094898978576776,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[361];split=linux_only;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":44},{"id":"pub-4dba25aacd5b5c1aeb82bd8f","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.09751802373446,"rawValue":0.4509751802373446,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[255];split=full/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":45},{"id":"pub-47597f43c60ae8fb66966cb5","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":54.91172987872457,"rawValue":0.5491172987872457,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[420];split=unlicensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":46},{"id":"pub-434567d01cf6458b719bb1fd","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":27.46547921284107,"rawValue":0.2746547921284107,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[299];split=licensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":47},{"id":"pub-3462e8d52cc5045b40d6f30f","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.25100182309368,"rawValue":0.39251001823093684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[79];split=full/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":48},{"id":"pub-319ca9d9335d5141c89ebe16","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":19.372652851069315,"rawValue":0.19372652851069314,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[124];split=full/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":49},{"id":"pub-315fcac0c284e1f683f4fa55","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.81714417182331,"rawValue":0.4681714417182331,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[372];split=linux_only;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":50},{"id":"pub-28dd1c20fe271e36c8c4ab41","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.31298337170319,"rawValue":0.4631298337170319,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[479];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":51},{"id":"pub-2035a311ce5767f49ab00325","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.575439428821085,"rawValue":0.46575439428821086,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[60];split=full/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":52},{"id":"pub-1e8cf832f752d7cf50010079","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":16.68727050234301,"rawValue":0.16687270502343013,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[544];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":53},{"id":"pub-1cffe73daaab4c7ce3c0dc29","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":54.736328337091045,"rawValue":0.5473632833709104,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[657];split=unlicensed/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":54},{"id":"pub-1aca0a8215634f0a39b1ffa4","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":19.829265389422996,"rawValue":0.19829265389422995,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[314];split=licensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":55},{"id":"pub-1101e0ece5b7337a9b0c7d34","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.65036485225495,"rawValue":0.44650364852254953,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[481];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":56},{"id":"pub-0d0179cdbce301b6928bb78c","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.124084089053454,"rawValue":0.4512408408905345,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[63];split=full/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":57},{"id":"pub-0cdd8e37df52e8e7ab565a1a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.81048909575152,"rawValue":0.4681048909575152,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[673];split=unlicensed/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":58},{"id":"pub-0508517d0785153b32b3bd5e","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":77.5482331051408,"rawValue":0.775482331051408,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[179];split=full/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":59},{"id":"pub-04e28f93128e4a300449ede9","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":19.420078029613762,"rawValue":0.19420078029613763,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[123];split=full/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":60},{"id":"pub-f895764d20cd3a330716ef45","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.57142857142857,"rawValue":0.2857142857142857,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[355];split=linux_only;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f62e026202d8d2ee4595bc52","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.57142857142857,"rawValue":0.2857142857142857,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[354];split=linux_only;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-f32d75b16f36cb62f65055d7","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":31.914893617021278,"rawValue":0.3191489361702128,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[421];split=unlicensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-ec5cf09f71b2354158ce7278","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[544];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-e847e60c1d6498af87c4be41","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":45.52238805970149,"rawValue":0.4552238805970149,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[178];split=full/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-e4f0a1356a56d8227f46b759","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.000000000000004,"rawValue":0.28,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[481];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-e3707e6372a46c838b5e08f6","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":30.592105263157894,"rawValue":0.3059210526315789,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[236];split=full/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-da24ea978abf610ff146a204","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.97872340425532,"rawValue":0.32978723404255317,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[656];split=unlicensed/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-d479de8e794b83ae7595adfd","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.413711583924346,"rawValue":0.2541371158392435,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[437];split=unlicensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-d1bd3444c747bd3e34b376b4","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":6.666666666666667,"rawValue":0.06666666666666667,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[539];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-cfa5c3d4bd3a2d7abab5c2d4","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.454545454545453,"rawValue":0.2545454545454545,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[63];split=full/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-ce6470e9ed5c7291d5d8b79c","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":31.914893617021278,"rawValue":0.3191489361702128,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[657];split=unlicensed/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-c81b11be284c735507c4ef16","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.495614035087712,"rawValue":0.29495614035087714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[240];split=full/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-c498b38a1ce4237bd457f515","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.605263157894733,"rawValue":0.29605263157894735,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[1];split=full;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-c33b188fe171f76d50fd30b7","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[79];split=full/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-c27705a3f3925820fafbdd40","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.142857142857142,"rawValue":0.2714285714285714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[359];split=linux_only;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-bb767c2e3231778adad6f5f9","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":47.76119402985074,"rawValue":0.47761194029850745,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[177];split=full/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-b34da02cd7cb5a7c92e4c583","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":30.592105263157894,"rawValue":0.3059210526315789,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[0];split=full;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-afe219a3a6bc4ba9c74ee4fa","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.605263157894733,"rawValue":0.29605263157894735,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[237];split=full/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-a6af3f1f744af718c9c8ddd0","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.97872340425532,"rawValue":0.32978723404255317,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[420];split=unlicensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-92c4b8e100b8420aa75711a3","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.27272727272727,"rawValue":0.2727272727272727,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[69];split=full/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-929f1a848d0d2a260e4b821a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.000000000000004,"rawValue":0.28,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[486];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-8bdf7df57274bdd28b1b5006","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.495614035087712,"rawValue":0.29495614035087714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[4];split=full;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-893f3988262f4489e52b8bed","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[498];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-842de8e3747a7a59b2ae7e44","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":47.01492537313433,"rawValue":0.4701492537313433,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[179];split=full/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":25},{"id":"pub-816d18872b7d8aadc73ed773","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":47.01492537313433,"rawValue":0.4701492537313433,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[597];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":26},{"id":"pub-80d594a868bf9a1177ee8ce3","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[302];split=licensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":27},{"id":"pub-7cf18b925909ef5a66788a31","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[137];split=full/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":28},{"id":"pub-7b4e3b3bd6663c9f46d82ba6","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.666666666666668,"rawValue":0.26666666666666666,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[361];split=linux_only;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":29},{"id":"pub-79dab6d53da10a079137d60f","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":47.76119402985074,"rawValue":0.47761194029850745,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[595];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":30},{"id":"pub-759440e050abaea460ce9b63","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":30.592105263157894,"rawValue":0.3059210526315789,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[3];split=full;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":31},{"id":"pub-758c55b3e6da0aeed4e52b8f","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":30.592105263157894,"rawValue":0.3059210526315789,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[239];split=full/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":32},{"id":"pub-6c77f44ede347361bbec808a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[314];split=licensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":33},{"id":"pub-698ac221ece4a0dfbb02dfc1","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[127];split=full/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":34},{"id":"pub-6762c9486816db6ce1a3d93a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":39.30348258706468,"rawValue":0.3930348258706468,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[197];split=full/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":35},{"id":"pub-63c70b76f6a3e807fbc8c612","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.903508771929822,"rawValue":0.2390350877192982,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[255];split=full/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":36},{"id":"pub-5fc96c43fa883dcd493c4701","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":30.0,"rawValue":0.3,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[479];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":37},{"id":"pub-55c1725f62054def4bc37ab8","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":30.0,"rawValue":0.3,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[60];split=full/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":38},{"id":"pub-471da0f64a0d655c030676f9","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":45.27363184079603,"rawValue":0.45273631840796025,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[599];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":39},{"id":"pub-46ffd099111ea9e726bd9dca","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":6.666666666666667,"rawValue":0.06666666666666667,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[542];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":40},{"id":"pub-3e6794a94397cd4524f8a75a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":31.7966903073286,"rawValue":0.317966903073286,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[422];split=unlicensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":41},{"id":"pub-3d6f9c1ff388ab470bf83579","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":5.263157894736842,"rawValue":0.05263157894736842,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[124];split=full/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":42},{"id":"pub-3d545d6a9d2bc0590d6f94c2","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.62411347517731,"rawValue":0.3262411347517731,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[654];split=unlicensed/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":43},{"id":"pub-3c412d2b144354e8146470ea","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.0,"rawValue":0.26,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[483];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":44},{"id":"pub-3aaafbe36f9219a52c74979a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":4.545454545454546,"rawValue":0.045454545454545456,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[303];split=licensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":45},{"id":"pub-39fe5f3894166737d64e201a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.27272727272727,"rawValue":0.2727272727272727,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[66];split=full/full-spectrum;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":46},{"id":"pub-3764df512e1ab415f3f4940d","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":5.0,"rawValue":0.05,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[543];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":47},{"id":"pub-342cad2e8ef529708e026ec8","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[299];split=licensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":48},{"id":"pub-2bd8a871e9d1ad9e5283c191","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.809523809523807,"rawValue":0.23809523809523808,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[372];split=linux_only;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":49},{"id":"pub-2a082d1e8d0a1cd1ea883219","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":39.30348258706468,"rawValue":0.3930348258706468,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[615];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":50},{"id":"pub-277d0172a09ee252995cf0cb","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.9473684210526314,"rawValue":0.039473684210526314,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[123];split=full/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":51},{"id":"pub-2328a6da5ad59618ab8776dc","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":5.263157894736842,"rawValue":0.05263157894736842,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[125];split=full/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":52},{"id":"pub-21faafb1bd8f95dfe090a718","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":45.52238805970149,"rawValue":0.4552238805970149,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[596];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":53},{"id":"pub-2055270cf8ce69a8de51ce65","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.413711583924346,"rawValue":0.2541371158392435,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[673];split=unlicensed/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":54},{"id":"pub-1d4a7af64bba55210d242ffc","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.903508771929822,"rawValue":0.2390350877192982,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[19];split=full;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":55},{"id":"pub-194bf68633c7e599d33c5c58","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":31.7966903073286,"rawValue":0.317966903073286,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[658];split=unlicensed/overall;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":56},{"id":"pub-167aed8182f4d7cde1e74e2a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[557];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":57},{"id":"pub-137460004930b032f20610a1","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":4.545454545454546,"rawValue":0.045454545454545456,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[305];split=licensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":58},{"id":"pub-0732bf78e98e035e3f5fccfa","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":45.27363184079603,"rawValue":0.45273631840796025,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[181];split=full/near-term;harness=codex;model=GPT-5.6-Sol;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":59},{"id":"pub-0652dbe86dd774c342b68bbc","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT-5.6-Sol","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.62411347517731,"rawValue":0.3262411347517731,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[418];split=unlicensed;harness=codex;model=GPT-5.6-Sol;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":60},{"id":"pub-fb3ad18a8cd2cae267ada7a9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.499288919011065,"rawValue":0.42499288919011063,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[685];split=unlicensed/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f5e4474e3c98d3f09a5f30d6","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.60911547618703,"rawValue":0.48609115476187026,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[433];split=unlicensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-f10f22c67db092be174db667","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":65.9506273356976,"rawValue":0.6595062733569761,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[198];split=full/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-e3329eb381e9357428fa2c4c","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":73.27376744045965,"rawValue":0.7327376744045965,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[602];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-ddcb4b1adeff6e1175de4615","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":27.28171107655029,"rawValue":0.2728171107655029,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[300];split=licensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-dd64555eeced27263a3dbeea","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.24908249653849,"rawValue":0.4024908249653849,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[32];split=full;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-dbce6538c4c4e95f19e6ecd1","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.38320139384285,"rawValue":0.4438320139384285,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[677];split=unlicensed/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-d969092417fd9cc65dfcbc21","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.77329832154835,"rawValue":0.5077329832154835,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[429];split=unlicensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-d3ef53f45e1a1b5699c043b4","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.33817997254725,"rawValue":0.4133817997254725,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[74];split=full/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-d2f73d6b6ef726926dd6ab59","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.9100891470296,"rawValue":0.439100891470296,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[380];split=linux_only;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-d14a4b4c36bdcdf025f475d9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.63069423114602,"rawValue":0.10630694231146019,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[561];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-cf7669554cb5a18f9f3e7edc","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.524049731525785,"rawValue":0.42524049731525787,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[384];split=linux_only;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-c5c2b8c378b3fc06d0c4f58c","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.511247258063579,"rawValue":0.10511247258063579,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[562];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-c4494571e871635c675c4abe","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.558362254691225,"rawValue":0.39558362254691226,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[493];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-c42a6dc4ae86352121a5d83d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.529184790268424,"rawValue":0.5252918479026842,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[426];split=unlicensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-c3c2dc7f0f2bee1dd76374eb","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":11.983054733814443,"rawValue":0.11983054733814444,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[139];split=full/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-c334e8d5e68ce6af99fe29f8","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.286059924850518,"rawValue":0.10286059924850519,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[565];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-bc1f1f09f5e54fd87cf6f9c5","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.12691704888627,"rawValue":0.4512691704888627,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[480];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-b8f8817b1cb44623bd1b24ae","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":73.87174887795719,"rawValue":0.7387174887795719,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[183];split=full/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-b8281423da7be4d5c9ebdfd9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.4942264470657,"rawValue":0.484942264470657,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[249];split=full/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-b056e6568c9b5c1adf36d4cf","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.566149659296249,"rawValue":0.1256614965929625,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[138];split=full/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-a986cddd51aa03b098ec1425","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.70206498203882,"rawValue":0.5070206498203882,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[244];split=full/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-a7eca400b795aad27d6bf954","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.914850377407577,"rawValue":0.14914850377407576,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[130];split=full/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-a4815a50df0a7628ad495a34","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":49.36421663563447,"rawValue":0.4936421663563447,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[365];split=linux_only;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-94b67fee73821b645a7817c5","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.24908249653849,"rawValue":0.4024908249653849,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[268];split=full/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":25},{"id":"pub-92715cae4adebbf89d40e101","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.121991039049476,"rawValue":0.5012199103904947,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[364];split=linux_only;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":26},{"id":"pub-8cd0ed5a0cd559077cf2c250","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.90348819796931,"rawValue":0.3390348819796931,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[510];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":27},{"id":"pub-8a0599eaa6db3ab9a31da57d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.4942264470657,"rawValue":0.484942264470657,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[13];split=full;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":28},{"id":"pub-8824f3bfad6abf6b1f564773","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":60.687436631111915,"rawValue":0.6068743663111191,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[626];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":29},{"id":"pub-856b36ef06c62e1c597fe0e7","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.18810640807843,"rawValue":0.46188106408078433,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[62];split=full/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":30},{"id":"pub-83e28f92b026a25e08a651f9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":70.84704112498615,"rawValue":0.7084704112498615,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[190];split=full/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":31},{"id":"pub-7973fe1458bfebb83f01ab94","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":70.84704112498615,"rawValue":0.7084704112498615,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[608];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":32},{"id":"pub-7737c2f17e9becdbaa0454c9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.48733134041199,"rawValue":0.4148733134041199,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[491];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":33},{"id":"pub-73e4a243871ac86438988e31","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":11.22775555549231,"rawValue":0.1122775555549231,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[142];split=full/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":34},{"id":"pub-729db638c65db81bdd630db0","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.91446401864527,"rawValue":0.3891446401864527,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[80];split=full/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":35},{"id":"pub-71e6c70b63dfeceb7cf2ea38","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":20.454787385590674,"rawValue":0.20454787385590675,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[313];split=licensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":36},{"id":"pub-655625e35603c38e247dcbbc","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.60911547618703,"rawValue":0.48609115476187026,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[669];split=unlicensed/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":37},{"id":"pub-637fe3966adcb838dee2a795","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.5716312064728,"rawValue":0.465716312064728,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[251];split=full/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":38},{"id":"pub-5cc41db365835d7e1920bbc2","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.99830316257406,"rawValue":0.3399830316257406,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[91];split=full/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":39},{"id":"pub-5a3466cda3046aceff2d5ab2","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.499288919011065,"rawValue":0.42499288919011063,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[449];split=unlicensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":40},{"id":"pub-57a08955edadd871c0ebc891","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.88643981010808,"rawValue":0.4688643981010808,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[371];split=linux_only;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":41},{"id":"pub-4c84754372ae77614744be12","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.77329832154835,"rawValue":0.5077329832154835,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[665];split=unlicensed/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":42},{"id":"pub-3b95ce3ff8225f8a8b08817f","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":15.022183083314491,"rawValue":0.15022183083314491,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[549];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":43},{"id":"pub-338c89b63dbdf38bcc0b010f","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":34.476830873934446,"rawValue":0.3447683087393445,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[89];split=full/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":44},{"id":"pub-316586f2c58c9f5eecd43f1f","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.70206498203882,"rawValue":0.5070206498203882,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[8];split=full;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":45},{"id":"pub-3061d3e66122660e084a7e92","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":19.554013940986383,"rawValue":0.19554013940986384,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[316];split=licensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":46},{"id":"pub-2f348743862ab0473cd45d3c","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.58635230185982,"rawValue":0.4258635230185982,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[22];split=full;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":47},{"id":"pub-2c0d8620b1c48194574f03f7","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":36.16544379079812,"rawValue":0.36165443790798124,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[506];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":48},{"id":"pub-2a135b10372e6e2207f027ba","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.38320139384285,"rawValue":0.4438320139384285,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[441];split=unlicensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":49},{"id":"pub-1dd9d46a2d07642cd5fa6917","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.5716312064728,"rawValue":0.465716312064728,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[15];split=full;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":50},{"id":"pub-1b9113fff5be28b105084db3","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":65.9506273356976,"rawValue":0.6595062733569761,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[616];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":51},{"id":"pub-172ef306bf8cf457eedd2b7e","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.58635230185982,"rawValue":0.4258635230185982,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[258];split=full/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":52},{"id":"pub-16d656cc4831906ea4ef291c","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.92923011990062,"rawValue":0.1092923011990062,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[145];split=full/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":53},{"id":"pub-168be6fe2c9738dd4190668f","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":60.687436631111915,"rawValue":0.6068743663111191,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[208];split=full/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":54},{"id":"pub-0ed4b513b44c6e98cc02ad13","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":19.280668783242483,"rawValue":0.19280668783242483,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[317];split=licensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":55},{"id":"pub-0dbd495c0e2682bbf7892d79","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":73.27376744045965,"rawValue":0.7327376744045965,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[184];split=full/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":56},{"id":"pub-0c0f54a6e837c062edbdcb17","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":11.405527444844456,"rawValue":0.11405527444844456,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[330];split=licensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":57},{"id":"pub-0ae28914a40ec444d4eb19ca","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.529184790268424,"rawValue":0.5252918479026842,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[662];split=unlicensed/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":58},{"id":"pub-0803b6400e46138d710b9660","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.081890647001854,"rawValue":0.10081890647001854,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[566];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":59},{"id":"pub-03b691c50737b10c840f985d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":73.87174887795719,"rawValue":0.7387174887795719,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[601];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":60},{"id":"pub-fd21f1bd049ab64c5b0bfdc0","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.01418439716312,"rawValue":0.2801418439716312,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[669];split=unlicensed/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fc5f03c01a67d66cc33ad3fa","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.181818181818183,"rawValue":0.18181818181818182,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[91];split=full/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-f834b4910101be362da1c29e","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":44.776119402985074,"rawValue":0.44776119402985076,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[183];split=full/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-f2158361cfd73fcfa6ca1a07","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.0,"rawValue":0.18,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[510];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-f0561a2c4cc03cd7c1e67c49","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.181818181818183,"rawValue":0.18181818181818182,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[89];split=full/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-ef81dfed78def04332ffd6fe","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.394736842105264,"rawValue":0.20394736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[268];split=full/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-eca9803be00acfe56d085b43","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.27272727272727,"rawValue":0.2727272727272727,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[62];split=full/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-e881f05190d1406d80e36085","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":41.7910447761194,"rawValue":0.417910447761194,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[184];split=full/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-e68f959d888c204e8177fbe7","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[138];split=full/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-e686587fd60fe5d903466558","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.71428571428571,"rawValue":0.2571428571428571,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[364];split=linux_only;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-e3404318923664885f967758","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":44.776119402985074,"rawValue":0.44776119402985076,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[601];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-e218f8d48192219d793e231e","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.476190476190474,"rawValue":0.20476190476190476,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[384];split=linux_only;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-df7fb2062845a2f5b50f4ab6","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.960526315789476,"rawValue":0.27960526315789475,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[244];split=full/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-df6bb4bc1d5b657cb0878491","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":38.80597014925373,"rawValue":0.3880597014925373,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[616];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-da9c3642f00c5849962ff2c4","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.08955223880597,"rawValue":0.3208955223880597,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[626];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-da311444a1dffe9859fbfff6","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":41.7910447761194,"rawValue":0.417910447761194,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[602];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-d397dde067c3660e51107095","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.090909090909092,"rawValue":0.09090909090909091,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[300];split=licensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-d2382428c1378b5b4d9c2b6d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.909090909090907,"rawValue":0.20909090909090908,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[80];split=full/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-d0c71eb437a0f832704538ea","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[566];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-be14d6e848b7743cea36771a","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.545454545454547,"rawValue":0.24545454545454545,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[74];split=full/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-ba6cdaa66f13162da4b0f52f","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.697368421052634,"rawValue":0.22697368421052633,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[22];split=full;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-ad471688defd5235b43d488b","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.631578947368425,"rawValue":0.27631578947368424,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[13];split=full;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-a99e505a878d6d5109a1b44a","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.960526315789476,"rawValue":0.27960526315789475,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[8];split=full;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-a6b8556ee6cdaf5e12c3fa31","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.78723404255319,"rawValue":0.2978723404255319,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[665];split=unlicensed/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-a097227f85b85976970aefed","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.78723404255319,"rawValue":0.2978723404255319,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[429];split=unlicensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":25},{"id":"pub-a08870fd2a3bf891ddb3a1a5","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.0,"rawValue":0.19,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[506];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":26},{"id":"pub-89d8be89b1da221b6a011819","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[549];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":27},{"id":"pub-88fd6d7e50bc2e2691ebac74","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.380952380952383,"rawValue":0.22380952380952382,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[380];split=linux_only;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":28},{"id":"pub-83471875135c750f4c3c00d1","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.08955223880597,"rawValue":0.3208955223880597,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[208];split=full/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":29},{"id":"pub-809efc2ec8f8db45fa6647de","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":38.80597014925373,"rawValue":0.3880597014925373,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[198];split=full/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":30},{"id":"pub-7f02b08a4a256d380c9f103d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[145];split=full/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":31},{"id":"pub-7e33262d68f680ff230c2e4b","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":43.28358208955223,"rawValue":0.43283582089552236,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[608];split=unlicensed/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":32},{"id":"pub-73b046f1385667539683a0cf","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[561];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":33},{"id":"pub-72cee62b6d5aa82474310790","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.432624113475175,"rawValue":0.29432624113475175,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[426];split=unlicensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":34},{"id":"pub-704106bb8449d393213aefd8","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[313];split=licensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":35},{"id":"pub-6dedb701e19f518e788d04f9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.46808510638298,"rawValue":0.24468085106382978,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[677];split=unlicensed/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":36},{"id":"pub-60cad6ab6f6ff56335ab7717","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.98581560283688,"rawValue":0.2198581560283688,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[449];split=unlicensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":37},{"id":"pub-557a5efa36c1be0ed7bd85fc","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.0,"rawValue":0.26,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[480];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":38},{"id":"pub-5500fc5bccda45c5cde29e50","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[565];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":39},{"id":"pub-537c39e651f8b8fc33839420","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.394736842105264,"rawValue":0.20394736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[32];split=full;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":40},{"id":"pub-51a203c61bbe644cbd99d944","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.631578947368425,"rawValue":0.27631578947368424,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[249];split=full/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":41},{"id":"pub-4dd3e406e619a27f438e97d9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.285714285714285,"rawValue":0.24285714285714285,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[371];split=linux_only;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":42},{"id":"pub-4c84d239ba739b856023aeca","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[316];split=licensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":43},{"id":"pub-4b02c4564e158fe113714e11","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.0,"rawValue":0.25,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[491];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":44},{"id":"pub-498aa5a5cc8b56f3a1c9521e","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[562];split=unlicensed/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":45},{"id":"pub-443f6ad4d2c2d062bef89a70","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[130];split=full/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":46},{"id":"pub-3d647fad3b4fada0b0bdc56a","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":43.28358208955223,"rawValue":0.43283582089552236,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[190];split=full/near-term;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":47},{"id":"pub-3c12ee6898e35883af449837","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.46808510638298,"rawValue":0.24468085106382978,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[441];split=unlicensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":48},{"id":"pub-33f3fa961cc6cc86114d6711","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.98581560283688,"rawValue":0.2198581560283688,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[685];split=unlicensed/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":49},{"id":"pub-2c0936b0c3e0660b351e58ae","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.986842105263158,"rawValue":0.2598684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[251];split=full/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":50},{"id":"pub-220f5993422521d17d57b9d6","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[317];split=licensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":51},{"id":"pub-213a96723659590e92633d30","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.0,"rawValue":0.22,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[493];split=unlicensed/full-spectrum;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":52},{"id":"pub-1a36d2853d62528c19c97050","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.697368421052634,"rawValue":0.22697368421052633,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[258];split=full/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":53},{"id":"pub-139e75eee6108a012ca3bd7d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[142];split=full/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":54},{"id":"pub-10ab209ba267daf06730387c","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.01418439716312,"rawValue":0.2801418439716312,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[433];split=unlicensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":55},{"id":"pub-0e6d9429769fe5caad2204e8","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[139];split=full/last-exam;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":56},{"id":"pub-0e0c84c5fcd251dc04c9d508","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.190476190476193,"rawValue":0.2619047619047619,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[365];split=linux_only;harness=codex;model=GPT-5.6-Terra;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":57},{"id":"pub-04acad77d0f42e20c8a82999","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.432624113475175,"rawValue":0.29432624113475175,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[662];split=unlicensed/overall;harness=codex;model=GPT-5.6-Terra;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-max","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":58},{"id":"pub-03eb51219eb529a7699701d4","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[330];split=licensed;harness=codex;model=GPT-5.6-Terra;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":59},{"id":"pub-0211dc49bc180c6210b9db4d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT-5.6-Terra","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.986842105263158,"rawValue":0.2598684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[15];split=full;harness=codex;model=GPT-5.6-Terra;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-high","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":60},{"id":"pub-fc5db40705b62cb133ef480d","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.70141224330407,"rawValue":0.4870141224330407,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[12];split=full;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e62cb31194b1292839b1a6ef","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":71.12863360362098,"rawValue":0.7112863360362098,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[607];split=unlicensed/near-term;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-dc584149dff40c7f2e855725","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.941419463910385,"rawValue":0.43941419463910386,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[680];split=unlicensed/overall;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-db63d2d33f3dbdbe9ec1daee","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":7.272727272727273,"rawValue":0.07272727272727274,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[337];split=licensed;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-dacb308f2f4610e3135820ae","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.00321393939394,"rawValue":0.4200321393939394,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[72];split=full/full-spectrum;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-c209825b34c489bbc1b42ce9","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.44353533333333,"rawValue":0.4044353533333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[492];split=unlicensed/full-spectrum;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-b5cfddbd4ec1a08f48b60e58","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.2877641079695,"rawValue":0.412877641079695,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[26];split=full;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-adf81a4204b484ab0d9514f6","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":16.495248496656565,"rawValue":0.16495248496656564,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[126];split=full/last-exam;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-a8b9256d4ec8f5c57e18725d","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":6.647333333333334,"rawValue":0.06647333333333334,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[580];split=unlicensed/last-exam;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-8cf4e9b5ed70881506f4c519","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.75886915040092,"rawValue":0.4575886915040092,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[374];split=linux_only;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-7cd2ced0be2c6c41c9da4aa0","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.941419463910385,"rawValue":0.43941419463910386,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[444];split=unlicensed;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-7951d220da5f9a2b3a9c1a47","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.450983770951275,"rawValue":0.4845098377095127,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[368];split=linux_only;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-7881459fc99581d460983a7a","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":16.393594595273655,"rawValue":0.16393594595273656,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[545];split=unlicensed/last-exam;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":13},{"id":"pub-648c8534aaf07f05a2c409c0","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.2877641079695,"rawValue":0.412877641079695,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[262];split=full/overall;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":14},{"id":"pub-6311802c2a90c3322c991aee","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":64.14472929875102,"rawValue":0.6414472929875101,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[617];split=unlicensed/near-term;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":15},{"id":"pub-507b3168305944757e0e8714","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":5.247894736842105,"rawValue":0.052478947368421056,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[163];split=full/last-exam;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":16},{"id":"pub-493a0b4b38d0d0f76fe6eb71","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.20995075154241,"rawValue":0.5020995075154241,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[431];split=unlicensed;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":17},{"id":"pub-2e99445131cd50bb13ee40e6","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.20995075154241,"rawValue":0.5020995075154241,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[667];split=unlicensed/overall;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":18},{"id":"pub-28d7c8a0e773b16159afb1ff","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":71.12863360362098,"rawValue":0.7112863360362098,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[189];split=full/near-term;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":19},{"id":"pub-264ca74cbc9a8e04933ff307","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":35.37496875263719,"rawValue":0.35374968752637187,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[87];split=full/full-spectrum;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":20},{"id":"pub-23979b88953963b0f01f7df4","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":37.312465627900906,"rawValue":0.37312465627900904,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[502];split=unlicensed/full-spectrum;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":21},{"id":"pub-2360d3dffb9156625513dff2","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":64.14472929875102,"rawValue":0.6414472929875101,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[199];split=full/near-term;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":22},{"id":"pub-218e4d132902de1050d7c9a1","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":29.36469136497635,"rawValue":0.2936469136497635,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[298];split=licensed;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":23},{"id":"pub-091a40ab5087035956ac297e","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.70141224330407,"rawValue":0.4870141224330407,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[248];split=full/overall;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":24},{"id":"pub-f22bc37454a1f03e253b069c","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[163];split=full/last-exam;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ef9d6c744650adde794dcaa9","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.0,"rawValue":0.24,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[492];split=unlicensed/full-spectrum;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ea78c40e4dbff3a51bbca76e","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[337];split=licensed;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-ea08c5397aa36f60cbf3e3e4","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.454545454545453,"rawValue":0.2545454545454545,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[72];split=full/full-spectrum;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-e75ffb0d5d696c9be27a2bbf","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.113475177304963,"rawValue":0.24113475177304963,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[680];split=unlicensed/overall;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-daa59ad914f70d7f944df34a","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.36842105263158,"rawValue":0.2236842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[262];split=full/overall;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-c3ef91575df831219e67b18e","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.657894736842106,"rawValue":0.2565789473684211,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[248];split=full/overall;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-bb88cb8af3e00615cc07311c","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":7.894736842105263,"rawValue":0.07894736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[126];split=full/last-exam;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-adc655fbfe8573bf1ba9e53f","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.95035460992908,"rawValue":0.2695035460992908,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[667];split=unlicensed/overall;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-913d2dbcf9c3bf2f26184cf1","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.36842105263158,"rawValue":0.2236842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[26];split=full;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-8f191c3dd1dd717d19299a33","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.657894736842106,"rawValue":0.2565789473684211,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[12];split=full;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-89ec55f67a3e6e7e3a66fdcb","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.95035460992908,"rawValue":0.2695035460992908,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[431];split=unlicensed;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-8385d5f48ce1c6482c0ddc7f","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.909090909090907,"rawValue":0.20909090909090908,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[87];split=full/full-spectrum;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":13},{"id":"pub-6ed23bc2eb2eb873a011da95","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.090909090909092,"rawValue":0.09090909090909091,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[298];split=licensed;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":14},{"id":"pub-5db3ff1096881ef3e76cc8f8","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[580];split=unlicensed/last-exam;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":15},{"id":"pub-55e3add333bbca37eec9494d","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.0,"rawValue":0.1,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[545];split=unlicensed/last-exam;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":16},{"id":"pub-49e576eb38f9aa8bc33cbc9c","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":35.07462686567165,"rawValue":0.35074626865671643,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[199];split=full/near-term;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":17},{"id":"pub-4275edd6a1c6a00447702fac","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.380952380952383,"rawValue":0.22380952380952382,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[374];split=linux_only;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":18},{"id":"pub-393ecf8640686598f8510bb7","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":37.3134328358209,"rawValue":0.373134328358209,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[189];split=full/near-term;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":19},{"id":"pub-1b5c05ee6c0bd5f7ae2b84c5","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.113475177304963,"rawValue":0.24113475177304963,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[444];split=unlicensed;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":20},{"id":"pub-10e35d8300079110fe6082f8","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.809523809523807,"rawValue":0.23809523809523808,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[368];split=linux_only;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":21},{"id":"pub-10689956bfacf31d2b02855e","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.0,"rawValue":0.23,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[502];split=unlicensed/full-spectrum;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":22},{"id":"pub-0a7ef4ada9d06512216e5dfe","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":35.07462686567165,"rawValue":0.35074626865671643,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[617];split=unlicensed/near-term;harness=claude_code;model=anthropic-claude-fable-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":23},{"id":"pub-09bbad213c828a339c6199e9","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"anthropic-claude-fable-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":37.3134328358209,"rawValue":0.373134328358209,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[607];split=unlicensed/near-term;harness=claude_code;model=anthropic-claude-fable-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-xhigh","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":24},{"id":"pub-fd8bf989d55c7b7e19c72803","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":60.450692410621734,"rawValue":0.6045069241062173,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[627];split=unlicensed/near-term;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fb6c9f79a9a54c39d21e4a45","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":61.712755225726845,"rawValue":0.6171275522572685,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[624];split=unlicensed/near-term;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-f710de4f65d5f894e9f14a49","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.856285924276156,"rawValue":0.10856285924276156,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[115];split=full/full-spectrum;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-f1f4935fea7cfcdac6502e94","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.90574125005675,"rawValue":0.3890574125005675,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[394];split=linux_only;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-ecd93dcacb1b60651f578cf9","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":21.56,"rawValue":0.21559999999999999,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[311];split=licensed;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-ea028351bca04bf8b06c7460","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":35.21506822590183,"rawValue":0.3521506822590183,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[41];split=full;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-e863798b8d422d3bba998e76","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":34.69339382470838,"rawValue":0.3469339382470838,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[397];split=linux_only;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-e50f7229dcb59148e3d08337","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":27.773197076864207,"rawValue":0.2777319707686421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[518];split=unlicensed/full-spectrum;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-e0dd9e3f41096f92d0ba963d","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":60.450692410621734,"rawValue":0.6045069241062173,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[209];split=full/near-term;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-dfa925c0bb276ef97190a980","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.029413579144805,"rawValue":0.08029413579144805,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[157];split=full/last-exam;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-ca3b8bb7c49d9ebefeb09a0a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.487183710893696,"rawValue":0.42487183710893694,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[23];split=full;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-c29049eaeb53d0fb883b4c93","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.029435149513931,"rawValue":0.10029435149513931,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[150];split=full/last-exam;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-c179e8c659923f013b52d9a0","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":9.501958811261163,"rawValue":0.09501958811261163,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[533];split=unlicensed/full-spectrum;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-bd57dc9d8252b7479add01a9","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.713736097477316,"rawValue":0.44713736097477313,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[439];split=unlicensed;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-b776397b72e4e7de4ebc3b25","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.26975564273962,"rawValue":0.4326975564273962,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[383];split=linux_only;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-b401b7d0e8ec93bc5927225a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.66383574955146,"rawValue":0.33663835749551463,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[698];split=unlicensed/overall;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-ab2bc7ea59d5870c5d778cc9","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":37.45618668551661,"rawValue":0.3745618668551661,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[694];split=unlicensed/overall;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-a0b111b7ff22e1b8c8b11bd5","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":35.80370011878202,"rawValue":0.35803700118782017,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[40];split=full;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-9ecdd6f20ba3da71c088478d","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.49605336355299,"rawValue":0.4049605336355299,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[267];split=full/overall;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-9cbfdcdeb8f995dbecd05548","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":57.99257416141243,"rawValue":0.5799257416141242,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[216];split=full/near-term;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-97ad07831e1d6afb344169dc","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":6.468889570193674,"rawValue":0.06468889570193674,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[340];split=licensed;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-96f79550e286c84814832685","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":36.51347439496799,"rawValue":0.36513474394967993,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[505];split=unlicensed/full-spectrum;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-8fd38db763813fd8fd9c4cec","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.68067137766155,"rawValue":0.45680671377661547,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[375];split=linux_only;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-847abf3e963dffb151adc68e","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.2929002531715,"rawValue":0.39292900253171503,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[497];split=unlicensed/full-spectrum;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-81464099b2aa310158c7a509","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":37.45618668551661,"rawValue":0.3745618668551661,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[458];split=unlicensed;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":25},{"id":"pub-7eb6bcaee41817435e338939","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":27.4703254822263,"rawValue":0.274703254822263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[101];split=full/full-spectrum;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":26},{"id":"pub-7e103d6a00daa5e55d001378","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":4.341299186851302,"rawValue":0.043412991868513016,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[166];split=full/last-exam;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":27},{"id":"pub-74898e0281548fd520c25d6b","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":37.22051359473687,"rawValue":0.3722051359473687,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[695];split=unlicensed/overall;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":28},{"id":"pub-71c358349b11f106a74c61b9","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":31.695780433940048,"rawValue":0.31695780433940046,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[280];split=full/overall;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":29},{"id":"pub-6e2a04a5c14e463fce17fe6b","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.531923866916753,"rawValue":0.08531923866916752,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[575];split=unlicensed/last-exam;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":30},{"id":"pub-6c11847911d9ccb96d782921","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.847049593392926,"rawValue":0.12847049593392926,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[554];split=unlicensed/last-exam;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":31},{"id":"pub-64f887844a4c590ebc6d3870","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.713736097477316,"rawValue":0.44713736097477313,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[675];split=unlicensed/overall;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":32},{"id":"pub-5b120107a614619d2581a6a8","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":17.642727381542535,"rawValue":0.17642727381542536,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[320];split=licensed;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":33},{"id":"pub-51f6999d97f8d80ca0eab429","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":7.863300882269293,"rawValue":0.07863300882269293,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[158];split=full/last-exam;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":34},{"id":"pub-516c2a107b925598f135e53b","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":29.108336390143062,"rawValue":0.29108336390143064,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[98];split=full/full-spectrum;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":35},{"id":"pub-4d593e6fdfae609ee5c90d05","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":27.546770029157376,"rawValue":0.27546770029157375,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[519];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":36},{"id":"pub-4aaa5b5a62667f60b8e3bc98","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":62.69364006481253,"rawValue":0.6269364006481253,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[620];split=unlicensed/near-term;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":37},{"id":"pub-46063b716c35c7adac808c00","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":6.4880043344759875,"rawValue":0.06488004334475987,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[338];split=licensed;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":38},{"id":"pub-4557038fef489e581a97f46e","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":61.712755225726845,"rawValue":0.6171275522572685,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[206];split=full/near-term;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":39},{"id":"pub-4425913ddbe87b1883f701f0","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.66383574955146,"rawValue":0.33663835749551463,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[462];split=unlicensed;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":40},{"id":"pub-421f4c4e093a1542a9a4a284","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":11.436091784257574,"rawValue":0.11436091784257574,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[140];split=full/last-exam;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":41},{"id":"pub-3afdafbe3b5b3f1f478adbf7","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":35.80370011878202,"rawValue":0.35803700118782017,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[276];split=full/overall;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":42},{"id":"pub-394fe4569e8c9e6e2c204d7e","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.739364141550475,"rawValue":0.38739364141550475,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[395];split=linux_only;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":43},{"id":"pub-37c3301ccc9c2ddf82382c9f","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":3.8603123033449824,"rawValue":0.03860312303344982,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[586];split=unlicensed/last-exam;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":44},{"id":"pub-30f90eb4047cd09e94c283cd","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":62.69364006481253,"rawValue":0.6269364006481253,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[202];split=full/near-term;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":45},{"id":"pub-2f358d40b96b1f2859a6df8c","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.487183710893696,"rawValue":0.42487183710893694,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[259];split=full/overall;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":46},{"id":"pub-2732077b4b795ea20ff74d50","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":57.99257416141243,"rawValue":0.5799257416141242,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[634];split=unlicensed/near-term;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":47},{"id":"pub-1e0cecc9706c28ddf11fb728","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.768284482818716,"rawValue":0.08768284482818715,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[573];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":48},{"id":"pub-1b7e78f663cbd5b748710d50","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.321514450874437,"rawValue":0.08321514450874437,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[576];split=unlicensed/last-exam;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":49},{"id":"pub-174845d1e43255181cd17438","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":37.61636630854757,"rawValue":0.37616366308547566,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[83];split=full/full-spectrum;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":50},{"id":"pub-15b865aa0617fe67148130d8","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":31.695780433940048,"rawValue":0.31695780433940046,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[44];split=full;harness=droid;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":51},{"id":"pub-1569ed04586695c65b80b489","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":36.61224944997091,"rawValue":0.3661224944997091,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[86];split=full/full-spectrum;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":52},{"id":"pub-143c853fc50c916a46ed203c","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":37.22051359473687,"rawValue":0.3722051359473687,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[459];split=unlicensed;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":53},{"id":"pub-1363bdc286239d81754ab36b","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":55.784441768979555,"rawValue":0.5578444176897955,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[218];split=full/near-term;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":54},{"id":"pub-127e9abde86dd3ee9a7608a1","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":55.784441768979555,"rawValue":0.5578444176897955,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[636];split=unlicensed/near-term;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":55},{"id":"pub-10601c5ee87f6b55327bf29a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":13.946830391958281,"rawValue":0.13946830391958281,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[325];split=licensed;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":56},{"id":"pub-0fe65268bde9c56bd939b67d","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.973334122411735,"rawValue":0.4197333412241174,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[687];split=unlicensed/overall;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":57},{"id":"pub-0db8e51118ed5752ea830daa","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":35.21506822590183,"rawValue":0.3521506822590183,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[277];split=full/overall;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":58},{"id":"pub-07cbddf456c798095db5ab9b","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.49605336355299,"rawValue":0.4049605336355299,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[31];split=full;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":59},{"id":"pub-01ed28738507d248a47372a4","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.973334122411735,"rawValue":0.4197333412241174,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[451];split=unlicensed;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":60},{"id":"pub-f6eb2640d9fe40b70dc85aea","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[576];split=unlicensed/last-exam;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f651c91fd62dc65e8309c7f6","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.421052631578945,"rawValue":0.18421052631578946,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[31];split=full;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-f640bfd24a49ea9dd1ab4722","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":15.789473684210526,"rawValue":0.15789473684210525,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[41];split=full;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-eb5a7a5c74a9e64c7403359a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":14.893617021276595,"rawValue":0.14893617021276595,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[459];split=unlicensed;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-e83a67a3b7db909b0fdd9e94","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[575];split=unlicensed/last-exam;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-e51541cf8797c662905e2ec3","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.69503546099291,"rawValue":0.22695035460992907,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[675];split=unlicensed/overall;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-e49605fa9a8a01dea438bfc9","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":14.539007092198581,"rawValue":0.1453900709219858,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[698];split=unlicensed/overall;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-de72143a8183168fb1cdbbe2","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[383];split=linux_only;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-d403adb4f58b52aa86fc0aa8","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":14.893617021276595,"rawValue":0.14893617021276595,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[695];split=unlicensed/overall;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-d10480624a6b1c957f7987a3","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.0,"rawValue":0.18,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[505];split=unlicensed/full-spectrum;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-d0448e8c2ea894e0b632e866","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":15.789473684210526,"rawValue":0.15789473684210525,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[277];split=full/overall;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-cc0e3c74147c769a97f8e49c","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.0,"rawValue":0.22,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[497];split=unlicensed/full-spectrum;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-c7d64699e1ac29ead11555fb","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[325];split=licensed;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-c70c396926c045f60397a483","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.727272727272727,"rawValue":0.12727272727272726,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[98];split=full/full-spectrum;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-c702ffa1e2ed9324310c5148","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.857142857142858,"rawValue":0.22857142857142856,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[375];split=linux_only;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-bf7a417cf0895a3b1d5c62c2","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":31.343283582089555,"rawValue":0.31343283582089554,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[202];split=full/near-term;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-bccf3d77219afd625a5e98c8","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":31.343283582089555,"rawValue":0.31343283582089554,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[620];split=unlicensed/near-term;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-b8e1f8410b495e66f8508f40","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.052631578947366,"rawValue":0.21052631578947367,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[259];split=full/overall;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-b5406f16fb5d4a00810d8ed0","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.148936170212767,"rawValue":0.19148936170212766,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[687];split=unlicensed/overall;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-ae99273287f738967794227a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.35820895522388,"rawValue":0.2835820895522388,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[216];split=full/near-term;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-ac7828957a7a04e4cf66b261","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[83];split=full/full-spectrum;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-a5eece0389dcfc883a01dba8","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.35820895522388,"rawValue":0.2835820895522388,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[634];split=unlicensed/near-term;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-a3c8f34bced9cedb5a21ed1f","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":15.238095238095239,"rawValue":0.1523809523809524,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[394];split=linux_only;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-a2c469b737f95178f259d67d","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[340];split=licensed;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-a13aaccb80dc4b728454acfd","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.181818181818183,"rawValue":0.18181818181818182,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[86];split=full/full-spectrum;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":25},{"id":"pub-9f9bafb52046ded5a83253ad","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":15.238095238095239,"rawValue":0.1523809523809524,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[397];split=linux_only;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":26},{"id":"pub-9c2d8eed4ace469777ff9c25","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.148936170212767,"rawValue":0.19148936170212766,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[451];split=unlicensed;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":27},{"id":"pub-97f21fc645d46e14aeab61f2","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[554];split=unlicensed/last-exam;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":28},{"id":"pub-91dc63b4c3ecdf4de711e876","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.35820895522388,"rawValue":0.2835820895522388,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[627];split=unlicensed/near-term;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":29},{"id":"pub-8f572318317059429b7a0205","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.0,"rawValue":0.1,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[518];split=unlicensed/full-spectrum;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":30},{"id":"pub-8b20358df211cc9592930198","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[150];split=full/last-exam;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":31},{"id":"pub-8692335995b2258ebad0e6e4","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[166];split=full/last-exam;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":32},{"id":"pub-7c53603edaf92598fa9eaabf","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[157];split=full/last-exam;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":33},{"id":"pub-79638ca316e175d916955eb1","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.35820895522388,"rawValue":0.2835820895522388,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[209];split=full/near-term;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":34},{"id":"pub-7330baf1face344817cc8b76","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.0,"rawValue":0.02,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[533];split=unlicensed/full-spectrum;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":35},{"id":"pub-6d3e816c62a65bd0e2e408ca","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.02127659574468,"rawValue":0.1702127659574468,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[458];split=unlicensed;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":36},{"id":"pub-613402515ff4fdcbe9480198","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[573];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":37},{"id":"pub-5e362706077dc9c8b06cff50","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.815789473684212,"rawValue":0.13815789473684212,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[40];split=full;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":38},{"id":"pub-5ca7c331a875df8ebefa471f","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.388059701492537,"rawValue":0.22388059701492538,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[218];split=full/near-term;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":39},{"id":"pub-595d6f524ffe1f0e38c51c09","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.388059701492537,"rawValue":0.22388059701492538,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[636];split=unlicensed/near-term;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":40},{"id":"pub-4e53835fafe728b45c85db53","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.142857142857142,"rawValue":0.17142857142857143,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[395];split=linux_only;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":41},{"id":"pub-3939a3d7774e775f676586e3","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.909090909090908,"rawValue":0.10909090909090909,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[101];split=full/full-spectrum;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":42},{"id":"pub-345b79e247067776fae30b66","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.090909090909092,"rawValue":0.09090909090909091,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[311];split=licensed;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":43},{"id":"pub-33ca4f9280179f7e3acebb0a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.02127659574468,"rawValue":0.1702127659574468,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[694];split=unlicensed/overall;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":44},{"id":"pub-31df4e2e809dad8cc87d56e2","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.1044776119403,"rawValue":0.291044776119403,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[624];split=unlicensed/near-term;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":45},{"id":"pub-3107ea084e44a17d6a0fca88","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.0,"rawValue":0.12,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[519];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":46},{"id":"pub-2bfc46cd50a785e393c870fb","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.486842105263158,"rawValue":0.13486842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[44];split=full;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":47},{"id":"pub-2b05d367e690bc666651a99a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.1044776119403,"rawValue":0.291044776119403,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[206];split=full/near-term;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":48},{"id":"pub-2a1b94c6c138963d62fae0a6","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[320];split=licensed;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":49},{"id":"pub-26b3af1db026d3c33bd918cc","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[140];split=full/last-exam;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":50},{"id":"pub-2628925667b577f1586924bc","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.486842105263158,"rawValue":0.13486842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[280];split=full/overall;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":51},{"id":"pub-222ae5bec5992bf5b3e9e073","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.69503546099291,"rawValue":0.22695035460992907,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[439];split=unlicensed;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":52},{"id":"pub-1d5dfd94ee563356a76531e3","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.052631578947366,"rawValue":0.21052631578947367,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[23];split=full;harness=cursor_cli;model=claude-opus-4-7;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"thinking-high","source_harness":"cursor_cli","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":53},{"id":"pub-1af492d09f16d9b05fedad7d","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":14.539007092198581,"rawValue":0.1453900709219858,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[462];split=unlicensed;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":54},{"id":"pub-15a59a95f176ed3f8b70a96e","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.815789473684212,"rawValue":0.13815789473684212,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[276];split=full/overall;harness=claude_code;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":55},{"id":"pub-14fa2e0270351adc95313b25","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[158];split=full/last-exam;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":56},{"id":"pub-14dfa4f99ed3da97a22e3d19","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.421052631578945,"rawValue":0.18421052631578946,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[267];split=full/overall;harness=ale_claw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":57},{"id":"pub-0f2f5d68c51cba3f827b74dd","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.6363636363636362,"rawValue":0.03636363636363636,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[115];split=full/full-spectrum;harness=droid;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":58},{"id":"pub-01adb0bd392af146215fb885","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[586];split=unlicensed/last-exam;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":59},{"id":"pub-00f48439416166692256052c","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[338];split=licensed;harness=openclaw;model=claude-opus-4-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":60},{"id":"pub-fd02b956f64b4f4b00d03b7d","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.189296504626746,"rawValue":0.44189296504626746,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[679];split=unlicensed/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fa3d4b7ff17b05ed7e112fbe","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.00608344770858,"rawValue":0.4300608344770858,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[681];split=unlicensed/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-f800508b3a85c417e03a9ea7","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.0,"rawValue":0.08,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[335];split=licensed;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-f2f4e5aca0db5b83f0c5dbf3","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.88552544602621,"rawValue":0.41885525446026206,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[489];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-ef2afcbf5b4dec6583c95715","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":59.8098576643749,"rawValue":0.598098576643749,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[213];split=full/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-eb7345bcb32cba01115e24d2","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.831305640791319,"rawValue":0.1083130564079132,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[331];split=licensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-e89379d9c5e013413f70423d","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.67685680771121,"rawValue":0.4067685680771121,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[690];split=unlicensed/overall;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-e527d501ae73a0b6ed4ab6c8","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.899846996079745,"rawValue":0.4089984699607975,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[390];split=linux_only;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-e1f543cbc22d0f7882bff161","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":62.73164650789494,"rawValue":0.6273164650789494,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[619];split=unlicensed/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-e1145ce6bfa874c11d2533f3","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.677645580102734,"rawValue":0.40677645580102734,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[266];split=full/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-db0552e1b08b92c7f03f214d","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.26526382013473,"rawValue":0.38265263820134726,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[82];split=full/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-d5c3797ab42eeba8a066e6c0","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.56096444480286,"rawValue":0.0856096444480286,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[574];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-d21c292c163a27d940e884cd","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":62.73164650789494,"rawValue":0.6273164650789494,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[201];split=full/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-cc3efd05f55bff8f1146ceaa","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":18.75675874992332,"rawValue":0.18756758749923322,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[318];split=licensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-c96a094e22e671eaa9c0aa57","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":28.733797491902628,"rawValue":0.2873379749190263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[99];split=full/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-ae8ccd9162636bbea853963e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.34878390395742,"rawValue":0.4234878390395742,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[24];split=full;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-a4394b9d8f99c66bbd4ffefa","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.431719138556748,"rawValue":0.10431719138556748,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[148];split=full/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-a4353cba2c92cade86963004","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.312084275574215,"rawValue":0.38312084275574215,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[37];split=full;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-a3dcf10d9e289695c08b888b","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":61.72996937063634,"rawValue":0.6172996937063634,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[205];split=full/near-term;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-a2580adb1510b85cfb9e1845","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.94146393591237,"rawValue":0.4694146393591237,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[672];split=unlicensed/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-9e214b6110d22d84bb80b783","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.189296504626746,"rawValue":0.44189296504626746,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[443];split=unlicensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-9c6e0574619ad7a163e4e9e6","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.14857743894153,"rawValue":0.4514857743894153,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[254];split=full/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-99a18c42b6f8a7369e700785","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":63.96884267777608,"rawValue":0.6396884267777608,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[200];split=full/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-9335da1296291a850210f2f7","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.94146393591237,"rawValue":0.4694146393591237,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[436];split=unlicensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-8fb41b7e0c4c451bac83bef7","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.14295886165538,"rawValue":0.42142958861655383,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[387];split=linux_only;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":25},{"id":"pub-899d81e62ee5606f35909d7c","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":37.93836589504913,"rawValue":0.3793836589504913,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[501];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":26},{"id":"pub-8322d64c35cec94e9039eba5","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":6.9505480733747005,"rawValue":0.069505480733747,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[160];split=full/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":27},{"id":"pub-6d252c1e6c3a94493bbf8537","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.40505369585742,"rawValue":0.1440505369585742,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[132];split=full/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":28},{"id":"pub-615eeb0233cb19bbb47ca78b","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":5.241315789473684,"rawValue":0.052413157894736835,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[164];split=full/last-exam;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":29},{"id":"pub-60cb28e26731d8df323b6feb","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.61128796135016,"rawValue":0.39611287961350156,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[76];split=full/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":30},{"id":"pub-5d03ca437284a8da918d17cd","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":61.72996937063634,"rawValue":0.6172996937063634,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[623];split=unlicensed/near-term;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":31},{"id":"pub-58204a5f86cac628ca3dfb96","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":59.8098576643749,"rawValue":0.598098576643749,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[631];split=unlicensed/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":32},{"id":"pub-546209e23c10f13bddc5ba2a","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.677645580102734,"rawValue":0.40677645580102734,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[30];split=full;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":33},{"id":"pub-520eb707155fa26c59056fbb","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":9.925073212365117,"rawValue":0.09925073212365117,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[567];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":34},{"id":"pub-4a990bb8c7a5756b05a3bcd2","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.335367385394136,"rawValue":0.3933536738539414,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[496];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":35},{"id":"pub-4a91e8593f90936acb639ff4","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.34878390395742,"rawValue":0.4234878390395742,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[260];split=full/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":36},{"id":"pub-49e76cb1785db950f66506ee","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":63.96884267777608,"rawValue":0.6396884267777608,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[618];split=unlicensed/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":37},{"id":"pub-3f6fb2c137523da5efd690ee","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.67685680771121,"rawValue":0.4067685680771121,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[454];split=unlicensed;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":38},{"id":"pub-3caad62dcf2ab1db17d2d0b7","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.06278923915453,"rawValue":0.4406278923915453,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[379];split=linux_only;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":39},{"id":"pub-31ac45ddbfe11b4ddda7caa5","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.312084275574215,"rawValue":0.38312084275574215,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[273];split=full/overall;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":40},{"id":"pub-317041e05913eb7147feb601","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.00608344770858,"rawValue":0.4300608344770858,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[445];split=unlicensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":41},{"id":"pub-2f466881d8321e221ba1a6b0","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":6.638999999999999,"rawValue":0.06638999999999999,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[581];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":42},{"id":"pub-20df2a57c0c8d2da49bb954b","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.14857743894153,"rawValue":0.4514857743894153,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[18];split=full;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":43},{"id":"pub-1e85b252dab80edab378bc6c","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.60039363451024,"rawValue":0.4460039363451024,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[377];split=linux_only;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":44},{"id":"pub-16ef764703fbf32568c3d319","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":29.847177241092893,"rawValue":0.2984717724109289,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[516];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":45},{"id":"pub-125e62636a10b0af78eb71a2","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":22.16703234140618,"rawValue":0.22167032341406182,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[308];split=licensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":46},{"id":"pub-0f737ad0795a98db4eafcbb1","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":16.051822822903798,"rawValue":0.16051822822903797,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[546];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":47},{"id":"pub-04fb8c1f2aa0ea40fd594cb2","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.132295860023824,"rawValue":0.4313229586002382,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[68];split=full/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":48},{"id":"pub-ffb14ef9a96458679d7d6771","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.113475177304963,"rawValue":0.24113475177304963,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[445];split=unlicensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fc735ea33ea3cd999fa1a6ac","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":35.82089552238806,"rawValue":0.3582089552238806,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[619];split=unlicensed/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-f68953aedee2c40676819644","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":16.447368421052634,"rawValue":0.16447368421052633,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[37];split=full;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-f1fe44666370853c84d854f9","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":37.3134328358209,"rawValue":0.373134328358209,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[631];split=unlicensed/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-e706c1c7f24f3bc33be1edb1","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[82];split=full/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-df949fa563bc4e10e761201e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.078014184397162,"rawValue":0.2907801418439716,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[672];split=unlicensed/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-dc7e35f68834bc985be07af0","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.0,"rawValue":0.24,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[489];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-d728065782f33ce56899afa1","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[567];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-d5b70b9805fab1c3023373e0","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.809523809523807,"rawValue":0.23809523809523808,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[377];split=linux_only;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-d2a2a557c23a3790fb3d7489","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.113475177304963,"rawValue":0.24113475177304963,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[681];split=unlicensed/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-c2e3be55d8fed62c7a7b0101","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[76];split=full/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-c14fcbd230ffa719469eb611","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[132];split=full/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-c0fd5e235cb823a5a0a5b59a","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[160];split=full/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-bd0e984dc37b5aa20d38d8d0","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[546];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-bc3cd3e1a34eaf4d28c590ba","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.142857142857142,"rawValue":0.17142857142857143,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[390];split=linux_only;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-ba6b9135c56c33d9b620946a","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.36842105263158,"rawValue":0.2236842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[266];split=full/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-b34cef5f99f204b3a2532d88","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[318];split=licensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-a635c865569adf94c2e76218","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.71428571428571,"rawValue":0.2571428571428571,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[379];split=linux_only;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-a2d883cc50915377b6024213","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":35.82089552238806,"rawValue":0.3582089552238806,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[201];split=full/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-a07c5918f3b25e7f2c5b8fa1","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":37.3134328358209,"rawValue":0.373134328358209,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[213];split=full/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-93c3b778bef21c9b87d2d364","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.0,"rawValue":0.12,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[516];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-8bcc47b27c004f5a84b7c0e3","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[501];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-8694a9a682dce135b3348483","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[581];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-8624d62b0cfa7472a187b0d6","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.73049645390071,"rawValue":0.1773049645390071,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[454];split=unlicensed;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-8176669573e383fb174e42f4","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[308];split=licensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":25},{"id":"pub-7bf7ea797dab6707e524b66a","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[164];split=full/last-exam;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":26},{"id":"pub-788619d572d29f12a85a13be","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.36842105263158,"rawValue":0.2236842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[30];split=full;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":27},{"id":"pub-780db29a95f67c809c2ca3f1","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.35820895522388,"rawValue":0.2835820895522388,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[205];split=full/near-term;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":28},{"id":"pub-754c807b6f586bc791ecf4f4","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.909090909090908,"rawValue":0.10909090909090909,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[99];split=full/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":29},{"id":"pub-73358c3bdcf6aff9d210c6bf","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[148];split=full/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":30},{"id":"pub-7260fa261c30a35c0bddccdf","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[496];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":31},{"id":"pub-71865a5a89f6ddc342c6620f","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[574];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":32},{"id":"pub-6d916245e5546af3aea6f1f9","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.73049645390071,"rawValue":0.1773049645390071,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[690];split=unlicensed/overall;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":33},{"id":"pub-6a4dbf8586278714e962f9b2","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.973684210526315,"rawValue":0.26973684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[254];split=full/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":34},{"id":"pub-6074c8d01555748b3fd9817f","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.36842105263158,"rawValue":0.2236842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[260];split=full/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":35},{"id":"pub-5b67a79bd65009e5d0b27814","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.113475177304963,"rawValue":0.24113475177304963,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[443];split=unlicensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":36},{"id":"pub-5136adf3983edf30ae25c6ab","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":43.28358208955223,"rawValue":0.43283582089552236,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[618];split=unlicensed/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":37},{"id":"pub-460ec2ecd65f7a97bcebe3fb","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":16.447368421052634,"rawValue":0.16447368421052633,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[273];split=full/overall;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":38},{"id":"pub-42a40fa455d4115a4e9e54b2","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[335];split=licensed;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":39},{"id":"pub-36d5da8d39be86be85350a72","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.636363636363637,"rawValue":0.23636363636363636,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[68];split=full/full-spectrum;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":40},{"id":"pub-2e5fbb9e93ffd3357e982912","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[387];split=linux_only;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":41},{"id":"pub-2b246f930d605522eab34d07","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[331];split=licensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-high","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":42},{"id":"pub-2ad970c925746cbc67a0af9e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.113475177304963,"rawValue":0.24113475177304963,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[679];split=unlicensed/overall;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":43},{"id":"pub-2611c5c9f703e2b727460b2c","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.973684210526315,"rawValue":0.26973684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[18];split=full;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":44},{"id":"pub-23a3cb2eeeea3a2c0f2f4832","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.36842105263158,"rawValue":0.2236842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[24];split=full;harness=claude_code;model=claude-opus-4-8;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":45},{"id":"pub-23385641e7db5367e04a9161","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.35820895522388,"rawValue":0.2835820895522388,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[623];split=unlicensed/near-term;harness=claude_code;model=claude-opus-4-8;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official+adj","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":null,"source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":46},{"id":"pub-0d90c50b6e602ad5569feb7e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":43.28358208955223,"rawValue":0.43283582089552236,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[200];split=full/near-term;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":47},{"id":"pub-0a4a8af126375116f7287329","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.078014184397162,"rawValue":0.2907801418439716,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[436];split=unlicensed;harness=claude_code;model=claude-opus-4-8;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"published+task-ratio","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":48},{"id":"pub-fc5a040bda11cbda66ca16b9","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":49.00101776002227,"rawValue":0.49001017760022275,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[247];split=full/overall;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e2d625ac29104675d8f1510f","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.54044140124046,"rawValue":0.4854044140124046,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[367];split=linux_only;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-cb599a9aad744f856f417424","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.404607194762896,"rawValue":0.504046071947629,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[430];split=unlicensed;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-b8125d8e1d992a9362d04163","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.488395328769904,"rawValue":0.39488395328769904,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[494];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-91fc912ecc129c75467e0255","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.404607194762896,"rawValue":0.504046071947629,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[666];split=unlicensed/overall;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-4ed8a0557259c78ed0541c2f","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":68.06900632972314,"rawValue":0.6806900632972314,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[610];split=unlicensed/near-term;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-3f96bcde836197240d2f9888","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":31.00955318743773,"rawValue":0.31009553187437733,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[297];split=licensed;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-3c95019adb2a0aee3a709b2f","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":68.06900632972314,"rawValue":0.6806900632972314,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[192];split=full/near-term;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-31fa40442a1ba39c2461af61","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.462177571609004,"rawValue":0.39462177571609003,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[77];split=full/full-spectrum;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-2d585483eaa995a5de4701b8","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.266092341932605,"rawValue":0.23266092341932604,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[120];split=full/last-exam;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-1c52caed82e6a9c3989a9592","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":24.633547464387465,"rawValue":0.24633547464387465,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[537];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-1b19721de4ffe82a84ede87a","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":49.00101776002227,"rawValue":0.49001017760022275,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[11];split=full;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-d9354f8c477430b2f5a9eb52","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.368794326241137,"rawValue":0.28368794326241137,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[666];split=unlicensed/overall;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b056d40d170aa4dd7c2fa4f6","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.0,"rawValue":0.1,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[537];split=unlicensed/last-exam;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-9c320dc2c280f605ac447efc","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.368794326241137,"rawValue":0.28368794326241137,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[430];split=unlicensed;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-8d2ba5f642c5059b648febce","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.818181818181817,"rawValue":0.21818181818181817,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[77];split=full/full-spectrum;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-8cc7383dc9a67a9c795d3991","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.973684210526315,"rawValue":0.26973684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[247];split=full/overall;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-86cfebca257f76251ae8406a","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":40.298507462686565,"rawValue":0.40298507462686567,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[610];split=unlicensed/near-term;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-642643582fa557bc02543966","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.090909090909092,"rawValue":0.09090909090909091,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[297];split=licensed;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-561d1ed4e4040f77fbb77b08","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.761904761904763,"rawValue":0.24761904761904763,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[367];split=linux_only;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-4cfd98e1dabf387f359d8857","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":7.894736842105263,"rawValue":0.07894736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[120];split=full/last-exam;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-1c73ba5764c7b6b1b0624102","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.973684210526315,"rawValue":0.26973684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[11];split=full;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-15da7fa7e9e32a9ef5298ddd","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.0,"rawValue":0.22,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[494];split=unlicensed/full-spectrum;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-0970201dd734938637ca3bfb","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":40.298507462686565,"rawValue":0.40298507462686567,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[192];split=full/near-term;harness=claude_code;model=claude-opus-5;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-abe5aec50a114ebf4afdfe38","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":28.997250926121826,"rawValue":0.28997250926121826,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[408];split=linux_only;harness=forgecode;model=claude-sonnet-4-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"forgecode","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6f72b75326fb1bd44ede427e","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":32.036709192716565,"rawValue":0.3203670919271656,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[403];split=linux_only;harness=hermes;model=claude-sonnet-4-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"hermes","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-6149775eb638501571fa335b","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":31.64026085498648,"rawValue":0.3164026085498648,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[404];split=linux_only;harness=openclaw;model=claude-sonnet-4-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-2a32534c04705c623ac670a6","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":30.942664987397716,"rawValue":0.30942664987397717,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[405];split=linux_only;harness=terminus_2;model=claude-sonnet-4-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"terminus_2","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-0ce78066f882ada7830faadb","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":19.77411110821487,"rawValue":0.19774111108214873,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[413];split=linux_only;harness=openhands;model=claude-sonnet-4-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openhands","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-dc71988a2c87829cdf2134fb","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.904761904761903,"rawValue":0.11904761904761904,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[405];split=linux_only;harness=terminus_2;model=claude-sonnet-4-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"terminus_2","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7607861a779bedfb5ffb6845","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.428571428571429,"rawValue":0.11428571428571428,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[404];split=linux_only;harness=openclaw;model=claude-sonnet-4-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-560024e5a6a2dc290450d9de","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.333333333333336,"rawValue":0.13333333333333336,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[408];split=linux_only;harness=forgecode;model=claude-sonnet-4-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"forgecode","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-32e756bdc29d8d4fedb30150","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.047619047619047,"rawValue":0.09047619047619047,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[413];split=linux_only;harness=openhands;model=claude-sonnet-4-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openhands","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-149c10e907f90ff68dc74e45","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.174603174603172,"rawValue":0.13174603174603172,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[403];split=linux_only;harness=hermes;model=claude-sonnet-4-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"official","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"hermes","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-f624435014a5daf8c9ca8fb5","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":3.193666666666666,"rawValue":0.03193666666666666,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[587];split=unlicensed/last-exam;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e5d252ef34f082ce1631a0df","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":27.62014846817647,"rawValue":0.2762014846817647,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[50];split=full;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-ccd35875c5ba3d1d8615cddc","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":27.62014846817647,"rawValue":0.2762014846817647,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[286];split=full/overall;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-c90d8921ee02ea1f10021c17","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.81171518153468,"rawValue":0.4381171518153468,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[645];split=unlicensed/near-term;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-c01406e5540c455c97336a5c","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":24.938953000000005,"rawValue":0.24938953000000005,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[522];split=unlicensed/full-spectrum;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-b8e792151be44ba282e526af","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":1.0909090909090908,"rawValue":0.010909090909090908,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[350];split=licensed;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-913956cc246d4f9ab7e1bc3b","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":29.68980544087109,"rawValue":0.2968980544087109,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[467];split=unlicensed;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-7978d1d20e6cdcb536fdb1f2","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.81171518153468,"rawValue":0.4381171518153468,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[227];split=full/near-term;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-5a5dc2e0c4413c719222e129","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.799048181818186,"rawValue":0.23799048181818186,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[105];split=full/full-spectrum;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-4b2e2d46d37505dd0b686417","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":29.68980544087109,"rawValue":0.2968980544087109,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[703];split=unlicensed/overall;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-4994bfa0282ed7f170e77aa4","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":30.84164065884846,"rawValue":0.30841640658848457,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[406];split=linux_only;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-2878b23a1287e6936c378a1e","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":2.521315789473684,"rawValue":0.02521315789473684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[169];split=full/last-exam;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-e39e88590c45ed0f3bb4c112","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.909090909090908,"rawValue":0.10909090909090909,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[105];split=full/full-spectrum;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ddd35ba0e9108c101d4c6125","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.0,"rawValue":0.11,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[522];split=unlicensed/full-spectrum;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-a442d5b09f7e1b497807bf36","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":14.126984126984127,"rawValue":0.14126984126984127,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[406];split=linux_only;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-9b3b3133b79d90e399b6197f","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.390350877192983,"rawValue":0.12390350877192982,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[50];split=full;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-98c56184313afece11d64719","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.900497512437813,"rawValue":0.19900497512437812,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[645];split=unlicensed/near-term;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-984212527a08ee769875e408","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[169];split=full/last-exam;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-727782f3fc986a819b4fee2a","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.356973995271867,"rawValue":0.13356973995271867,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[467];split=unlicensed;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-52503a690a5794ff4021191c","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[587];split=unlicensed/last-exam;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-38d17b92c542d3d820b14837","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.356973995271867,"rawValue":0.13356973995271867,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[703];split=unlicensed/overall;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-3106487ce24c6e5bbae9b2b7","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.900497512437813,"rawValue":0.19900497512437812,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[227];split=full/near-term;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-1bdfe5862e023d438f65fdcc","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.390350877192983,"rawValue":0.12390350877192982,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[286];split=full/overall;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-0594c26811c2a2f4558a5968","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[350];split=licensed;harness=openclaw;model=deepseek-v4-pro;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-e7853f957319839f915f625e","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":34.805421707762456,"rawValue":0.3480542170776246,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[460];split=unlicensed;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e69af8736b1cdfc5180e81b5","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":31.130806689351527,"rawValue":0.31130806689351526,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[701];split=unlicensed/overall;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ca78c576d48d33ec655329fb","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":32.70766092627965,"rawValue":0.3270766092627965,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[278];split=full/overall;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-c97e533c5001e7f3961d5d37","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":49.540530993013434,"rawValue":0.4954053099301343,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[221];split=full/near-term;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-be239a7475ed1697ce8c9336","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":37.17262877253931,"rawValue":0.3717262877253931,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[396];split=linux_only;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-bcb00d1d18ada9dd7083149c","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":31.130806689351527,"rawValue":0.31130806689351526,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[465];split=unlicensed;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-b60dd410b21cd44aec5e5c04","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":29.27265620525372,"rawValue":0.2927265620525372,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[47];split=full;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-b42cd2aa17066a153978ee64","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":32.70766092627965,"rawValue":0.3270766092627965,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[42];split=full;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-a79037db55cf6334f953b0d2","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":3.868333333333333,"rawValue":0.03868333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[585];split=unlicensed/last-exam;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-a6d01db319fe00efbf44be86","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":0.9123684210526317,"rawValue":0.009123684210526317,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[175];split=full/last-exam;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-a5e5058ff7dd5ac49be97caf","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":54.94600574739002,"rawValue":0.5494600574739001,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[219];split=full/near-term;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-7beb12704b2b91f7d347be94","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":1.1556666666666666,"rawValue":0.011556666666666666,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[593];split=unlicensed/last-exam;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-7671e5368715653164119a29","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":25.745041514387506,"rawValue":0.25745041514387507,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[521];split=unlicensed/full-spectrum;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":13},{"id":"pub-728389d5d6295a092647ab7d","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":5.454545454545454,"rawValue":0.05454545454545454,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[343];split=licensed;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":14},{"id":"pub-6c5afd12ef0583d93fa4d856","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":22.743563333333334,"rawValue":0.22743563333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[526];split=unlicensed/full-spectrum;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":15},{"id":"pub-68f98fd60f091a336e1f8687","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":29.27265620525372,"rawValue":0.2927265620525372,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[283];split=full/overall;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":16},{"id":"pub-48852245cfd4101a46429e8d","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":49.540530993013434,"rawValue":0.4954053099301343,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[639];split=unlicensed/near-term;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":17},{"id":"pub-3483bc5f8f49a8c302bf61f2","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.585057575757578,"rawValue":0.23585057575757576,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[106];split=full/full-spectrum;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":18},{"id":"pub-2c5c4014a7ae58adedc3e59e","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":34.805421707762456,"rawValue":0.3480542170776246,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[696];split=unlicensed/overall;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":19},{"id":"pub-246767e928c6b7258d12fed4","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":54.94600574739002,"rawValue":0.5494600574739001,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[637];split=unlicensed/near-term;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":20},{"id":"pub-235a970e2ff493fe249b1a62","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":5.818181818181818,"rawValue":0.05818181818181818,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[342];split=licensed;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":21},{"id":"pub-2132661f65104389aa66e3df","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":32.4334461581786,"rawValue":0.32433446158178597,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[402];split=linux_only;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":22},{"id":"pub-205f79d357f9c2cdafc4bcc3","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":26.386401376715913,"rawValue":0.2638640137671591,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[102];split=full/full-spectrum;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":23},{"id":"pub-05d758475e58520b0d258242","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":3.0539473684210527,"rawValue":0.03053947368421053,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[168];split=full/last-exam;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":24},{"id":"pub-f9d68efcb1210b8e06d00e25","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":16.447368421052634,"rawValue":0.16447368421052633,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[42];split=full;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f452cb53123922e2955d92b2","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.35820895522388,"rawValue":0.2835820895522388,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[637];split=unlicensed/near-term;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-e7f7018558abe22b973f459e","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.095238095238095,"rawValue":0.18095238095238095,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[396];split=linux_only;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-e63f1f538bea6cfda8021577","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.727272727272727,"rawValue":0.12727272727272726,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[102];split=full/full-spectrum;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-e076ef161b65e1517455b0f4","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.119402985074625,"rawValue":0.26119402985074625,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[639];split=unlicensed/near-term;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-e057ff66bda5eab5b440f0ea","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":14.144736842105262,"rawValue":0.14144736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[47];split=full;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-c5d6b421f86639a4daec1ac1","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[168];split=full/last-exam;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-c14214b235a37059a9d56988","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":15.24822695035461,"rawValue":0.1524822695035461,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[701];split=unlicensed/overall;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-b75bd0668a154a011eef633f","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.119402985074625,"rawValue":0.26119402985074625,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[221];split=full/near-term;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-b0b4584bf655643036ddc71f","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[342];split=licensed;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-839cbfc491754402bab38cad","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.73049645390071,"rawValue":0.1773049645390071,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[696];split=unlicensed/overall;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-829f95511808de24fd62f02c","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[593];split=unlicensed/last-exam;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-7842baf0bc705093b76183b1","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.0,"rawValue":0.12,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[521];split=unlicensed/full-spectrum;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":13},{"id":"pub-72cf7bef3a379c5672e19904","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.35820895522388,"rawValue":0.2835820895522388,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[219];split=full/near-term;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":14},{"id":"pub-6f6079110983475f228c5d80","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.909090909090908,"rawValue":0.10909090909090909,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[106];split=full/full-spectrum;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":15},{"id":"pub-62227228620041cc65b504ac","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":16.447368421052634,"rawValue":0.16447368421052633,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[278];split=full/overall;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":16},{"id":"pub-5dee03e5e25d74f54e377b3e","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[175];split=full/last-exam;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":17},{"id":"pub-5118ee368b8df0edf000d6ad","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.73049645390071,"rawValue":0.1773049645390071,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[460];split=unlicensed;harness=gemini_cli;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"gemini_cli","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":18},{"id":"pub-4d1bdde6802ee352942329d9","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[585];split=unlicensed/last-exam;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":19},{"id":"pub-2ffbe01a995441bc468d87f4","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":15.714285714285714,"rawValue":0.15714285714285714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[402];split=linux_only;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":20},{"id":"pub-2c8bcd72013b959dd6e2d900","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":14.144736842105262,"rawValue":0.14144736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[283];split=full/overall;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":21},{"id":"pub-2418b220e0720060d4bbcc60","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[343];split=licensed;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":22},{"id":"pub-1ab3bd4d2caaff3e2e65d47d","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.0,"rawValue":0.1,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[526];split=unlicensed/full-spectrum;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":23},{"id":"pub-03321c3ec83d7c3521f89247","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3-1-pro-preview","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":15.24822695035461,"rawValue":0.1524822695035461,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[465];split=unlicensed;harness=openclaw;model=gemini-3-1-pro-preview;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+adj","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":24},{"id":"pub-eb80142978847acd89ca6619","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":28.13736456137783,"rawValue":0.2813736456137783,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[49];split=full;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8a9a0cc13f64a7150fd86c25","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":6.1951272828356405,"rawValue":0.061951272828356406,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[582];split=unlicensed/last-exam;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-855c3efa4f9a0d26d99748aa","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":28.13736456137783,"rawValue":0.2813736456137783,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[285];split=full/overall;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-6630b63a2534be4e4822f6dc","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":30.781676019478986,"rawValue":0.3078167601947899,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[407];split=linux_only;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-6376d9eb394d8d3c67b65e29","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":6.184574170659716,"rawValue":0.061845741706597165,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[162];split=full/last-exam;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-5ad4eff9674f4ae8a3e9dd7d","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.56096373519291,"rawValue":0.45560963735192905,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[643];split=unlicensed/near-term;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-542ed245aae98817af4e5c30","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.56096373519291,"rawValue":0.45560963735192905,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[225];split=full/near-term;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-3a29bacf1a04aebb3715cfb8","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":21.708428055845587,"rawValue":0.21708428055845586,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[109];split=full/full-spectrum;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-1ecab2421971ed7a768dda57","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":22.985225925256938,"rawValue":0.22985225925256939,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[525];split=unlicensed/full-spectrum;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-1a7a129d83367aa4981c1bca","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":29.66678841504093,"rawValue":0.2966678841504093,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[468];split=unlicensed;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-18de558107c6dd9019ceb285","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":29.66678841504093,"rawValue":0.2966678841504093,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[704];split=unlicensed/overall;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-0231b03d2be25442af8313c9","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.532931528060011,"rawValue":0.08532931528060012,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[333];split=licensed;harness=openclaw;model=glm-5-1;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-e45a3ccbc8a7c686fdee5085","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.513157894736842,"rawValue":0.11513157894736842,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[285];split=full/overall;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b2c66f7a2b874e4bcfcc3f98","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.090909090909092,"rawValue":0.09090909090909091,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[109];split=full/full-spectrum;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-9e75c220c8bc6c4dc860bd40","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.411347517730496,"rawValue":0.12411347517730496,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[704];split=unlicensed/overall;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-9e52055c6fdb0257301d258e","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[333];split=licensed;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-8a50d3dd085cfcae7f241250","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.149253731343283,"rawValue":0.20149253731343283,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[643];split=unlicensed/near-term;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-78ee6b147c0cf0c1b57bd9f9","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.513157894736842,"rawValue":0.11513157894736842,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[49];split=full;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-71c308349ae50235503a7549","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[162];split=full/last-exam;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-5a623c1f2386d27d7e68b8bb","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[582];split=unlicensed/last-exam;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-597cd37f27e539a9b6b151cc","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.857142857142856,"rawValue":0.12857142857142856,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[407];split=linux_only;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-4dc778f7336067df0e81528b","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.149253731343283,"rawValue":0.20149253731343283,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[225];split=full/near-term;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-4b23f2a467dcd1dc6808e6c8","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.411347517730496,"rawValue":0.12411347517730496,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[468];split=unlicensed;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-1017cde10f66caf84aa23eb2","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5-1","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.0,"rawValue":0.1,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[525];split=unlicensed/full-spectrum;harness=openclaw;model=glm-5-1;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-f5a7998db59d0a7f53713cf7","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.15980756094993,"rawValue":0.4415980756094993,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[378];split=linux_only;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c13ab23dea3c83a21eb55f9d","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.08540680857251,"rawValue":0.41085406808572505,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[264];split=full/overall;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-a6f5954ef4fdf8ddf3fd8b35","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.93016937371446,"rawValue":0.4293016937371446,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[446];split=unlicensed;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-86e511628c668f98956422df","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.965036002828908,"rawValue":0.12965036002828909,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[553];split=unlicensed/last-exam;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-86b7297ba52b15aefd0fdeb6","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":17.438904837207417,"rawValue":0.17438904837207417,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[321];split=licensed;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-81eb6a50abbf2c2556281ac5","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.08540680857251,"rawValue":0.41085406808572505,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[28];split=full;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-76310f4b0d00775d5a61f56c","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":36.974012794692094,"rawValue":0.36974012794692096,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[85];split=full/full-spectrum;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-7628c82ab13fdc06af896c59","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.862606139319707,"rawValue":0.12862606139319707,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[135];split=full/last-exam;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-4d825d81b1f66cd7f08cdfec","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":60.33126287817978,"rawValue":0.6033126287817978,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[628];split=unlicensed/near-term;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-2861afe20e464a4fa94f26cf","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":36.83141407416131,"rawValue":0.36831414074161306,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[504];split=unlicensed/full-spectrum;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-219b3cc71bcda401d5cf88f0","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.93016937371446,"rawValue":0.4293016937371446,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[682];split=unlicensed/overall;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-1bd2b2ed2032cd8b58ea6274","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":60.33126287817978,"rawValue":0.6033126287817978,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[210];split=full/near-term;harness=claude_code;model=glm-5-2;variant=max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-fba8a3365033e223c3eefdc8","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[321];split=licensed;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ece8ac165ff96318da5cb5df","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.98581560283688,"rawValue":0.2198581560283688,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[682];split=unlicensed/overall;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ceb45305b7e4ad3cc460ea19","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.394736842105264,"rawValue":0.20394736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[28];split=full;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b51a88d957082b0f6af8ef4a","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.83582089552239,"rawValue":0.3283582089552239,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[628];split=unlicensed/near-term;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-b0fbbd82d997aef2389f2729","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.83582089552239,"rawValue":0.3283582089552239,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[210];split=full/near-term;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-88caa6a764cc9aa82fd01d18","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.0,"rawValue":0.18,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[504];split=unlicensed/full-spectrum;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-84e27dc61096ded1a59adab5","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[135];split=full/last-exam;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-3d3cc9280d742d4e8b0df11a","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[553];split=unlicensed/last-exam;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-3caa7d09414aaa949c067317","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.394736842105264,"rawValue":0.20394736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[264];split=full/overall;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-32f5513bc1dd696e0d93d3e7","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.809523809523807,"rawValue":0.23809523809523808,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[378];split=linux_only;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-23e8b1d0b66f03bcef36527b","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.98581560283688,"rawValue":0.2198581560283688,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[446];split=unlicensed;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-183c430d607d06af0fb0d633","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5-2","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.181818181818183,"rawValue":0.18181818181818182,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[85];split=full/full-spectrum;harness=claude_code;model=glm-5-2;variant=max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"max","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-f75f854e88cb0b68ed42eadb","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":13.3367903588527,"rawValue":0.133367903588527,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[293];split=full/overall;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e99a1a50e53d55d4187bc1cc","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.134244963505957,"rawValue":0.08134244963505957,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[156];split=full/last-exam;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-e79e204361e9113a5fdc4181","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.84820514506088,"rawValue":0.4184820514506088,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[388];split=linux_only;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-e21fc38dd28c0835cf203ed3","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":34.34147041110992,"rawValue":0.3434147041110992,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[90];split=full/full-spectrum;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-df3452ecb14ddf8a79300f29","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.97438926549261,"rawValue":0.23974389265492613,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[234];split=full/near-term;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-dd9df60aff8e056a8ee797dd","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":13.116963841633533,"rawValue":0.13116963841633533,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[417];split=linux_only;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-dba4b47ff595df54c1caa64f","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":22.512809050649008,"rawValue":0.22512809050649008,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[527];split=unlicensed/full-spectrum;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-d9cc1e70f94f35dad0191ed1","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.97438926549261,"rawValue":0.23974389265492613,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[652];split=unlicensed/near-term;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-ca4466f40f4c890bfbca564a","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.373662346275324,"rawValue":0.45373662346275323,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[226];split=full/near-term;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-bbdbfdfc39d685ae856ba6e2","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.81077761612696,"rawValue":0.33810777616126964,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[400];split=linux_only;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-ad41805c01ce80c6bce61364","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":4.169525455925167,"rawValue":0.041695254559251674,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[348];split=licensed;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-ab91d540281ec06f82dc7e09","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":57.8007915138166,"rawValue":0.578007915138166,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[635];split=unlicensed/near-term;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-a4613cd1feaa3790ed62f901","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":57.8007915138166,"rawValue":0.578007915138166,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[217];split=full/near-term;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":13},{"id":"pub-8d8f8fc07a2282e5fb97d237","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":13.781504500323477,"rawValue":0.13781504500323477,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[711];split=unlicensed/overall;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":14},{"id":"pub-8d0a7dbd83792c2b18c564f8","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":13.3367903588527,"rawValue":0.133367903588527,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[57];split=full;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":15},{"id":"pub-74d9d10b2239337d2ae9f01e","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":37.32607831603129,"rawValue":0.3732607831603129,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[38];split=full;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":16},{"id":"pub-6f65a58ce6eab3b6b3829589","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[594];split=unlicensed/last-exam;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":17},{"id":"pub-6b1d72c49587a54b0b80448e","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":13.781504500323477,"rawValue":0.13781504500323477,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[475];split=unlicensed;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":18},{"id":"pub-6621c9971a7e5c5bcb06e306","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":28.63832387982169,"rawValue":0.2863832387982169,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[48];split=full;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":19},{"id":"pub-5619a46a1b5ef63137314d8d","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.303376953774212,"rawValue":0.10303376953774213,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[564];split=unlicensed/last-exam;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":20},{"id":"pub-53998c78b891fc4077654743","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.912759744833906,"rawValue":0.39912759744833903,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[692];split=unlicensed/overall;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":21},{"id":"pub-4fd09e5278d8a4f83b6806fd","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.373662346275324,"rawValue":0.45373662346275323,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[644];split=unlicensed/near-term;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":22},{"id":"pub-4b3a285c5a0e75fe9b707b5d","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.912759744833906,"rawValue":0.39912759744833903,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[456];split=unlicensed;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":23},{"id":"pub-268361a7e21530403db0b820","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":3.2727272727272725,"rawValue":0.03272727272727272,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[349];split=licensed;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":24},{"id":"pub-2521c0d59e4257929a205015","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":30.617200210871605,"rawValue":0.30617200210871603,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[466];split=unlicensed;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":25},{"id":"pub-1cb2250052a0adfe6445a184","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":22.93891731877182,"rawValue":0.22938917318771823,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[107];split=full/full-spectrum;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":26},{"id":"pub-1c19b3cd822fe07c2ecc27f4","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":7.636363636363636,"rawValue":0.07636363636363636,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[336];split=licensed;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":27},{"id":"pub-16feba65573dfdffb50afa18","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":7.318078210639797,"rawValue":0.07318078210639797,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[159];split=full/last-exam;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":28},{"id":"pub-13e915170f768dd13f09ca9e","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":28.63832387982169,"rawValue":0.2863832387982169,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[284];split=full/overall;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":29},{"id":"pub-11f76f47625caf8eb47653d1","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":30.617200210871605,"rawValue":0.30617200210871603,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[702];split=unlicensed/overall;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":30},{"id":"pub-105eb85459e33883ea1096a4","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":9.269565733477076,"rawValue":0.09269565733477077,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[570];split=unlicensed/last-exam;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":31},{"id":"pub-0c6b2cedb00c508d2b33f504","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":7.652873704683727,"rawValue":0.07652873704683727,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[117];split=full/full-spectrum;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":32},{"id":"pub-0a10d85ce71fd6bc50841ad3","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":6.7381610751520995,"rawValue":0.06738161075152099,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[535];split=unlicensed/full-spectrum;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":33},{"id":"pub-077dc0f94f71fe086524af8c","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":34.85832185191737,"rawValue":0.3485832185191737,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[508];split=unlicensed/full-spectrum;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":34},{"id":"pub-04958cf1223e94ce8d86f9c0","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":37.32607831603129,"rawValue":0.3732607831603129,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[274];split=full/overall;harness=openclaw;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":35},{"id":"pub-01aabe3a203ab920ae33eb44","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[176];split=full/last-exam;harness=codex;model=gpt-5-4;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":36},{"id":"pub-fdc3c3c41705ae8235eb50e5","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":7.236842105263158,"rawValue":0.07236842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[57];split=full;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fc4820c233a58428d9c3d982","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[570];split=unlicensed/last-exam;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-fc3c4b2c088ab654a0bae493","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[348];split=licensed;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-f743f455a463aed1a7c7b501","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[176];split=full/last-exam;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-e8435619ab415c1e5e9fec20","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[159];split=full/last-exam;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-e1137040f1f6d490868fec8b","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.432835820895523,"rawValue":0.13432835820895522,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[652];split=unlicensed/near-term;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-c6b7074014a87ce756bcc901","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[336];split=licensed;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-bab5088e748db298a31df020","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[564];split=unlicensed/last-exam;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-b33c1658fe6d6a198fcd5a1d","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.10401891252955,"rawValue":0.22104018912529552,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[456];split=unlicensed;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-b20dbb5a0d109cc8b3d24d7a","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.393939393939394,"rawValue":0.19393939393939394,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[90];split=full/full-spectrum;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-adf338934b7ec7c747d07122","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.76595744680851,"rawValue":0.1276595744680851,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[702];split=unlicensed/overall;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-ad7dceccfe3e9926e6499c7f","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":4.0,"rawValue":0.04,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[535];split=unlicensed/full-spectrum;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-a90cc8ebc6165e3f3e2ab674","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[156];split=full/last-exam;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":13},{"id":"pub-9c56b2b0ec7480e3afde3485","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.76595744680851,"rawValue":0.1276595744680851,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[466];split=unlicensed;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":14},{"id":"pub-9bc2109cb2621fd9692a4325","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.6363636363636362,"rawValue":0.03636363636363636,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[117];split=full/full-spectrum;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":15},{"id":"pub-957e26d2e38c99765fd55892","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":33.582089552238806,"rawValue":0.3358208955223881,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[217];split=full/near-term;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":16},{"id":"pub-946fd2c714e0d996833cd67c","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.50438596491228,"rawValue":0.2050438596491228,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[274];split=full/overall;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":17},{"id":"pub-8eba3601636670a9c717ae7d","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.842105263157894,"rawValue":0.11842105263157894,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[48];split=full;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":18},{"id":"pub-88f2e49f1bbf05bac5ef5962","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.8955223880597,"rawValue":0.208955223880597,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[644];split=unlicensed/near-term;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":19},{"id":"pub-8889bb686f6e198a8d9e0001","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.53968253968254,"rawValue":0.2253968253968254,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[388];split=linux_only;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":20},{"id":"pub-81225fa42011bd0b99b9d61f","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":33.582089552238806,"rawValue":0.3358208955223881,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[635];split=unlicensed/near-term;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":21},{"id":"pub-72dde664fba631aa02df911c","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.333333333333334,"rawValue":0.13333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[400];split=linux_only;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":22},{"id":"pub-72607593fd6cd9434bd72042","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.10401891252955,"rawValue":0.22104018912529552,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[692];split=unlicensed/overall;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":23},{"id":"pub-7145684ff69c5626c18ac174","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":7.801418439716312,"rawValue":0.07801418439716312,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[711];split=unlicensed/overall;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":24},{"id":"pub-5cae43f7817a652400fbd72c","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.8955223880597,"rawValue":0.208955223880597,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[226];split=full/near-term;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":25},{"id":"pub-522467c21e4ad8b983d6e719","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":8.0,"rawValue":0.08,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[527];split=unlicensed/full-spectrum;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":26},{"id":"pub-4d2e9e146c54530b799e5a1e","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[594];split=unlicensed/last-exam;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":27},{"id":"pub-47c83f827dc25ca188d4f69c","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":6.666666666666667,"rawValue":0.06666666666666667,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[417];split=linux_only;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":28},{"id":"pub-4232e5da5de7ded6d39e7dd2","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.50438596491228,"rawValue":0.2050438596491228,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[38];split=full;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":29},{"id":"pub-3d25436e9aa3ea245b9f7944","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.090909090909092,"rawValue":0.09090909090909091,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[107];split=full/full-spectrum;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":30},{"id":"pub-3c171c58a8b5f7a545fea448","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.333333333333332,"rawValue":0.19333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[508];split=unlicensed/full-spectrum;harness=openclaw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":31},{"id":"pub-3bc968352e5710c9512917cb","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.842105263157894,"rawValue":0.11842105263157894,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[284];split=full/overall;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":32},{"id":"pub-2ce4d62093b087c195e224fe","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":7.801418439716312,"rawValue":0.07801418439716312,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[475];split=unlicensed;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":33},{"id":"pub-2296de3e6bbc2b1169e4dc04","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.432835820895523,"rawValue":0.13432835820895522,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[234];split=full/near-term;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":34},{"id":"pub-12244711ac2239c4f7a6a675","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[349];split=licensed;harness=ale_claw;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":35},{"id":"pub-01ef282874cf49f1e64ed870","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5-4","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":7.236842105263158,"rawValue":0.07236842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[293];split=full/overall;harness=codex;model=gpt-5-4;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":36},{"id":"pub-fc565824dc206c4c30070ebf","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.0514071141378,"rawValue":0.410514071141378,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[75];split=full/full-spectrum;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f2d8c37031b04e6cbe1b8103","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":21.759244620262603,"rawValue":0.21759244620262602,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[310];split=licensed;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ef075b67f125b20d83bfca67","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":32.66237832837682,"rawValue":0.3266237832837682,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[512];split=unlicensed/full-spectrum;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-eea3bdaabea73b15905440c7","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.98827476250321,"rawValue":0.33988274762503207,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[92];split=full/full-spectrum;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-eb82ed0df83c5105d0318805","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.85130493292262,"rawValue":0.42851304932922624,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[447];split=unlicensed;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-eb2a916e0f8f45a619abcf79","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":13.388631862928818,"rawValue":0.13388631862928818,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[134];split=full/last-exam;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-e99c8276e03ac13b2cf0111e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":22.245515154818698,"rawValue":0.22245515154818696,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[307];split=licensed;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-e7271d47f8c7262cd2ebddb1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":62.28643923694662,"rawValue":0.6228643923694662,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[204];split=full/near-term;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-e5e00d7ca31e4bf321855dc6","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":67.36635047623743,"rawValue":0.6736635047623744,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[194];split=full/near-term;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-e5b420cb37337c2295d94c11","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.85130493292262,"rawValue":0.42851304932922624,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[683];split=unlicensed/overall;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-e1e1474536100d13e298f942","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":47.931391024591825,"rawValue":0.47931391024591824,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[250];split=full/overall;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-e1824372df728e8b48e057a1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":53.53736949071799,"rawValue":0.5353736949071799,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[220];split=full/near-term;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-e010e3e6467895c173c4f88b","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":34.037980623548364,"rawValue":0.34037980623548364,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[509];split=unlicensed/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":13},{"id":"pub-da1c358e07fdee0f65911e42","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":67.0526220421849,"rawValue":0.670526220421849,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[614];split=unlicensed/near-term;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":14},{"id":"pub-d939cdbb4efdf80f51448a54","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":22.151974870087496,"rawValue":0.22151974870087496,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[309];split=licensed;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":15},{"id":"pub-d5332267a38fc59c01057542","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":49.86758940421861,"rawValue":0.49867589404218615,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[668];split=unlicensed/overall;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":16},{"id":"pub-d373ed3ffa206ca708fbd1c5","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.24577544679435,"rawValue":0.44245775446794344,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[678];split=unlicensed/overall;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":17},{"id":"pub-d3334f1c81b648c6e8e0779d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.052577733234465,"rawValue":0.4605257773323447,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[373];split=linux_only;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":18},{"id":"pub-d2f7dac353aef25697087f12","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":36.369651146549295,"rawValue":0.3636965114654929,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[39];split=full;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":19},{"id":"pub-d0f2695fd76ae814674e3bea","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.072396960513174,"rawValue":0.42072396960513175,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[25];split=full;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":20},{"id":"pub-ce19c29f6eb70093573e9a7c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.61494333333333,"rawValue":0.3361494333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[511];split=unlicensed/full-spectrum;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":21},{"id":"pub-cc49b0e5c6460c7b9c6812ec","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.24577544679435,"rawValue":0.44245775446794344,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[442];split=unlicensed;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":22},{"id":"pub-cbdbd2cc0626cbeb0d68ce30","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.815059202118556,"rawValue":0.12815059202118556,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[136];split=full/last-exam;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":23},{"id":"pub-cb347015ced34a9b3308dfc1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.28945809408762,"rawValue":0.3828945809408762,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[457];split=unlicensed;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":24},{"id":"pub-ca2eccd36602eec862317ef1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":67.1923973370024,"rawValue":0.6719239733700241,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[613];split=unlicensed/near-term;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":25},{"id":"pub-c8a0974f156b98b8d8cf406e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.715784362305094,"rawValue":0.44715784362305094,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[256];split=full/overall;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":26},{"id":"pub-c8998e48c3365a4dbaca0e91","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":58.1507678464108,"rawValue":0.581507678464108,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[633];split=unlicensed/near-term;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":27},{"id":"pub-be908b3641a52f6ee26e4a71","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":9.88031738224873,"rawValue":0.09880317382248731,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[152];split=full/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":28},{"id":"pub-bb5f3d1d2d93c33762880940","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.14434329247684,"rawValue":0.4114434329247684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[689];split=unlicensed/overall;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":29},{"id":"pub-b9f4928ba8c4728e50e4459a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.77666494040948,"rawValue":0.39776664940409484,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[33];split=full;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":30},{"id":"pub-b3bcee673822acca459254ef","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.213636363636361,"rawValue":0.14213636363636362,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[324];split=licensed;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":31},{"id":"pub-b1f0033dd8f63ed0d789b611","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.78443243363864,"rawValue":0.4578443243363864,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[253];split=full/overall;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":32},{"id":"pub-b07dca9cf28b15acb0ebb134","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":13.36927248629258,"rawValue":0.1336927248629258,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[552];split=unlicensed/last-exam;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":33},{"id":"pub-b025afdec8c791b96a2e38f7","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":16.0932671435888,"rawValue":0.160932671435888,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[128];split=full/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":34},{"id":"pub-ae811d30a9ffa294c9d99893","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":58.1507678464108,"rawValue":0.581507678464108,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[215];split=full/near-term;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":35},{"id":"pub-ad4476988e7382c15e1adf5d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.427017991504975,"rawValue":0.43427017991504974,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[382];split=linux_only;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":36},{"id":"pub-aa740a90b817f0e6a1b98274","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":67.36635047623743,"rawValue":0.6736635047623744,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[612];split=unlicensed/near-term;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":37},{"id":"pub-aa0c92796633712391d3331c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.246512581474754,"rawValue":0.3924651258147475,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[35];split=full;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":38},{"id":"pub-a9ca4179d5fd4e8205132634","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":35.194903030447946,"rawValue":0.3519490303044795,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[88];split=full/full-spectrum;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":39},{"id":"pub-a832ec95c99b22ea0feb6462","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":59.268098867917075,"rawValue":0.5926809886791707,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[632];split=unlicensed/near-term;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":40},{"id":"pub-a7ad23ff8cbe268e70f7a290","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":47.65873786588782,"rawValue":0.47658737865887824,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[671];split=unlicensed/overall;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":41},{"id":"pub-a64b9a4613387d74f42bb39e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.3952530257971,"rawValue":0.333952530257971,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[93];split=full/full-spectrum;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":42},{"id":"pub-a3ecce0130acb4e60461c004","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":35.034393333492744,"rawValue":0.3503439333349274,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[507];split=unlicensed/full-spectrum;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":43},{"id":"pub-a3c47bea75d7dfd6766d18bd","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.68007855798093,"rawValue":0.40680078557980925,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[29];split=full;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":44},{"id":"pub-a20892d3c24f2cc04be2ab4f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.732247368421053,"rawValue":0.10732247368421054,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[147];split=full/last-exam;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":45},{"id":"pub-a1cfe013d8d12c66703d4375","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.476081556733405,"rawValue":0.46476081556733406,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[438];split=unlicensed;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":46},{"id":"pub-a07525cfcf82d4be04630f63","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":9.288872936875796,"rawValue":0.09288872936875796,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[569];split=unlicensed/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":47},{"id":"pub-9f7da013f5d7cf4962ed5107","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.68007855798093,"rawValue":0.40680078557980925,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[265];split=full/overall;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":48},{"id":"pub-993d13d5042229c922b9a26c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.874277625057212,"rawValue":0.10874277625057213,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[146];split=full/last-exam;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":49},{"id":"pub-9786bc00f990308b8969f794","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":62.28643923694662,"rawValue":0.6228643923694662,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[622];split=unlicensed/near-term;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":50},{"id":"pub-9658eacaaaf67bfe2f9d93de","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.36390008783094,"rawValue":0.4136390008783094,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[688];split=unlicensed/overall;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":51},{"id":"pub-90c0735c799326910ef19e53","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":53.53736949071799,"rawValue":0.5353736949071799,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[638];split=unlicensed/near-term;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":52},{"id":"pub-89489ae7d7bca8d13ad06dc0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.3010500664481,"rawValue":0.33301050066448096,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[94];split=full/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":53},{"id":"pub-847187b6c2d3cca8b0b2f6f3","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":15.026470889226193,"rawValue":0.15026470889226193,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[548];split=unlicensed/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":54},{"id":"pub-7c1077a2645dc284afafce93","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.28945809408762,"rawValue":0.3828945809408762,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[693];split=unlicensed/overall;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":55},{"id":"pub-77d8d257fb7142d42a172d56","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":72.4007875487875,"rawValue":0.724007875487875,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[185];split=full/near-term;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":56},{"id":"pub-6e30d4de3230168f57e122e1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":11.836513333333333,"rawValue":0.11836513333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[558];split=unlicensed/last-exam;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":57},{"id":"pub-6d622e162da2a7611b2ee2f6","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.697865037281922,"rawValue":0.12697865037281922,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[555];split=unlicensed/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":58},{"id":"pub-6d215d3cbcec9c4a7b2649a4","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.112848158466555,"rawValue":0.23112848158466556,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[306];split=licensed;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":59},{"id":"pub-685d2dfbd2c416895fd68fa5","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.715784362305094,"rawValue":0.44715784362305094,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[20];split=full;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":60},{"id":"pub-67395833f34c76d4d02825dc","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.0849060167781,"rawValue":0.480849060167781,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[369];split=linux_only;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":61},{"id":"pub-60fd5883a1ecbb94badb0689","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":40.481898174638836,"rawValue":0.40481898174638836,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[391];split=linux_only;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":62},{"id":"pub-5bd3eaf432ceee898bec7943","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.60896833299899,"rawValue":0.4560896833299899,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[376];split=linux_only;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":63},{"id":"pub-59408b70dfdec344d586a525","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":49.235206278939025,"rawValue":0.49235206278939025,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[366];split=linux_only;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":64},{"id":"pub-53d2960aa70c69ffd58df20a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.59374165601684,"rawValue":0.1459374165601684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[550];split=unlicensed/last-exam;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":65},{"id":"pub-4fa188896103a74bbd885186","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":32.401937123043226,"rawValue":0.3240193712304323,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[95];split=full/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":66},{"id":"pub-4ad2b722e09839908163d27a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.072396960513174,"rawValue":0.42072396960513175,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[261];split=full/overall;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":67},{"id":"pub-48e9dd4e236fe81537ef219c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.3527140090938,"rawValue":0.39352714009093803,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[495];split=unlicensed/full-spectrum;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":68},{"id":"pub-46efa6cabb57523cf8b879a7","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.246512581474754,"rawValue":0.3924651258147475,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[271];split=full/overall;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":69},{"id":"pub-4349c5562fb8394110ae4058","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":49.86758940421861,"rawValue":0.49867589404218615,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[432];split=unlicensed;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":70},{"id":"pub-43107790d0dcb20cc0ca0caf","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.848904115546352,"rawValue":0.12848904115546353,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[326];split=licensed;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":71},{"id":"pub-3a110876fb958828ea90538c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.105454545454544,"rawValue":0.12105454545454544,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[328];split=licensed;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":72},{"id":"pub-37b6c0a32dc19a273165b913","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.77666494040948,"rawValue":0.39776664940409484,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[269];split=full/overall;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":73},{"id":"pub-36e1fdb29f84c2a177e03c03","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":67.0526220421849,"rawValue":0.670526220421849,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[196];split=full/near-term;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":74},{"id":"pub-3491ee0833a58c79482ab351","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":36.93544973841823,"rawValue":0.3693544973841823,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[503];split=unlicensed/full-spectrum;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":75},{"id":"pub-326eda4ad21c6cbe04a307c9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":47.65873786588782,"rawValue":0.47658737865887824,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[435];split=unlicensed;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":76},{"id":"pub-316ba157d064ff1aa53a1f5b","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.192087907608547,"rawValue":0.14192087907608547,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[133];split=full/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":77},{"id":"pub-304400ef8073190cd815bfa4","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":47.931391024591825,"rawValue":0.47931391024591824,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[14];split=full;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":78},{"id":"pub-2e84b87e59fd1291f06568c4","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":37.55080680526102,"rawValue":0.37550806805261017,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[84];split=full/full-spectrum;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":79},{"id":"pub-289ee322f4a6093cfe963be8","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.4105932064375,"rawValue":0.394105932064375,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[393];split=linux_only;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":80},{"id":"pub-2791969c8cf7fd0e447bb903","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":36.369651146549295,"rawValue":0.3636965114654929,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[275];split=full/overall;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":81},{"id":"pub-26286e2aee16f2f2a1e4d430","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.70750484411592,"rawValue":0.38707504844115914,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[81];split=full/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":82},{"id":"pub-257d9f9b7892708bddb30d47","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":38.70842922925664,"rawValue":0.38708429229256636,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[500];split=unlicensed/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":83},{"id":"pub-2553ba97da2df15f4cefa6d8","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":10.429084991739138,"rawValue":0.10429084991739138,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[563];split=unlicensed/last-exam;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":84},{"id":"pub-22e775fbae3f14ea8143c736","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":67.1923973370024,"rawValue":0.6719239733700241,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[195];split=full/near-term;harness=openclaw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":85},{"id":"pub-225711d4cc34199ea9b25b38","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":11.761216637194266,"rawValue":0.11761216637194266,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[329];split=licensed;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":86},{"id":"pub-1f88fc986786115d372d7f23","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":11.26774697160647,"rawValue":0.11267746971606471,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[141];split=full/last-exam;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":87},{"id":"pub-1ec1800b15c8b48608f97bb9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":12.633812830701533,"rawValue":0.12633812830701532,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[556];split=unlicensed/last-exam;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":88},{"id":"pub-1def2740b2343ccd2997b14a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":39.445811616626685,"rawValue":0.39445811616626686,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[392];split=linux_only;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":89},{"id":"pub-18c86b79af6037739db8c53d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":59.268098867917075,"rawValue":0.5926809886791707,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[214];split=full/near-term;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":90},{"id":"pub-0b2977efa8de5ca48b31b649","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.14434329247684,"rawValue":0.4114434329247684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[453];split=unlicensed;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":91},{"id":"pub-08e0acc6c3cc3b7ef21f5bbb","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":72.4007875487875,"rawValue":0.724007875487875,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[603];split=unlicensed/near-term;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":92},{"id":"pub-08643266fe930eee203f4498","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":31.952009526446133,"rawValue":0.31952009526446135,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[513];split=unlicensed/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":93},{"id":"pub-07b50e24d631ba0b65ee92e9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":45.78443243363864,"rawValue":0.4578443243363864,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[17];split=full;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":94},{"id":"pub-06e86c03872b64266b75aa8f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":41.36390008783094,"rawValue":0.4136390008783094,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[452];split=unlicensed;harness=droid;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":95},{"id":"pub-02ef8d1b6421b2a04ad01321","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.476081556733405,"rawValue":0.46476081556733406,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[674];split=unlicensed/overall;harness=codex;model=gpt-5-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":96},{"id":"pub-fce6996ee67e86c216eb2cd4","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.026315789473685,"rawValue":0.23026315789473684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[253];split=full/overall;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fcdcd941ca8a1895414f6de8","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[550];split=unlicensed/last-exam;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-f41b2aaa506c77b828aa0b0a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.38157894736842,"rawValue":0.2138157894736842,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[265];split=full/overall;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-f4132774ab0e329028b5f945","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.0,"rawValue":0.22,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[500];split=unlicensed/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-f2732c8b866e47d0107d4e94","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.761904761904763,"rawValue":0.24761904761904763,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[373];split=linux_only;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-f223d4144bca6a8473f1b64f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[326];split=licensed;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-f1d5290218f1132234715a69","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[569];split=unlicensed/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-ee080f0ed3e7ca4bc0028053","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.45175438596491,"rawValue":0.2445175438596491,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[20];split=full;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-ecd29bc62962ed49318884a5","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.0,"rawValue":0.18,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[507];split=unlicensed/full-spectrum;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-ecc351c9c0ca0e60bee1fb8f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.858156028368796,"rawValue":0.19858156028368795,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[689];split=unlicensed/overall;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-ec6ca73b3b0ba8d0a756aba3","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.736842105263158,"rawValue":0.19736842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[35];split=full;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-e8070661264628aa3f8a8b60","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[511];split=unlicensed/full-spectrum;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-dfb2649b1aa4e04b6227ad31","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":37.3134328358209,"rawValue":0.373134328358209,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[195];split=full/near-term;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":13},{"id":"pub-df13e19b9a986580d7455ae8","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.439716312056735,"rawValue":0.18439716312056736,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[457];split=unlicensed;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":14},{"id":"pub-d54b74a2f88990f433939008","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.12293144208038,"rawValue":0.26122931442080377,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[674];split=unlicensed/overall;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":15},{"id":"pub-d1eb9bad409eb23f821c06fe","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[558];split=unlicensed/last-exam;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":16},{"id":"pub-d110cb2776dc5bd242f59474","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.37313432835821,"rawValue":0.2537313432835821,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[220];split=full/near-term;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":17},{"id":"pub-cf1deb7a8c426fd06bc0c4ff","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.105263157894736,"rawValue":0.17105263157894737,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[39];split=full;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":18},{"id":"pub-cbbee6012c9cc10b2e74b8e9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.27659574468085,"rawValue":0.2127659574468085,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[688];split=unlicensed/overall;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":19},{"id":"pub-ca97e36e7323b8d536821d1a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":1.3157894736842104,"rawValue":0.013157894736842105,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[128];split=full/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":20},{"id":"pub-c65ee792ac6e40bb242a8bab","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[563];split=unlicensed/last-exam;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":21},{"id":"pub-c157fa305e90a49af4efe819","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[329];split=licensed;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":22},{"id":"pub-c13bde3ef9b7c380f96ab7fe","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.65079365079365,"rawValue":0.2365079365079365,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[376];split=linux_only;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":23},{"id":"pub-c0fa1dbb7cd03af315811d48","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":44.02985074626866,"rawValue":0.44029850746268656,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[185];split=full/near-term;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":24},{"id":"pub-bc9a4c8a2edc8995efef6370","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.818181818181817,"rawValue":0.21818181818181817,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[81];split=full/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":25},{"id":"pub-bc00f187903bda5d025071ac","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.113475177304963,"rawValue":0.24113475177304963,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[671];split=unlicensed/overall;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":26},{"id":"pub-b75c09c2d2970689e357e73d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[92];split=full/full-spectrum;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":27},{"id":"pub-b1835335aaa28832c62d6218","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.27659574468085,"rawValue":0.2127659574468085,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[452];split=unlicensed;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":28},{"id":"pub-aaaf3e845d3325433b796dc0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.850746268656714,"rawValue":0.29850746268656714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[633];split=unlicensed/near-term;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":29},{"id":"pub-aa5a3f0c7b09524a3711c3fa","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":1.6666666666666667,"rawValue":0.016666666666666666,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[548];split=unlicensed/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":30},{"id":"pub-a633dfbad563cf02812e6c14","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.12293144208038,"rawValue":0.26122931442080377,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[438];split=unlicensed;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":31},{"id":"pub-a5860f9c64380bcb01d17646","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.37313432835821,"rawValue":0.2537313432835821,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[638];split=unlicensed/near-term;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":32},{"id":"pub-a38f6b8664f9035c51c0026f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.04964539007092,"rawValue":0.23049645390070922,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[447];split=unlicensed;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":33},{"id":"pub-9d0fad82f3deabbaa0bb077d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.421052631578945,"rawValue":0.18421052631578946,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[269];split=full/overall;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":34},{"id":"pub-9cd60112dd96cb6fbc263bac","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[146];split=full/last-exam;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":35},{"id":"pub-9ad00d7aaed95e611fdaa292","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.83582089552239,"rawValue":0.3283582089552239,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[612];split=unlicensed/near-term;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":36},{"id":"pub-9807312bcf277c5c90b34e69","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[141];split=full/last-exam;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":37},{"id":"pub-943e66b37aa28ba384aaa870","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.710526315789476,"rawValue":0.21710526315789475,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[25];split=full;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":38},{"id":"pub-93396e57f940b443ff037d31","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.710526315789476,"rawValue":0.21710526315789475,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[261];split=full/overall;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":39},{"id":"pub-8f5806c0311bfee761c0bd22","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.666666666666668,"rawValue":0.22666666666666668,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[503];split=unlicensed/full-spectrum;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":40},{"id":"pub-8b780fce7a03ff3e8793fcc0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.850746268656714,"rawValue":0.29850746268656714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[215];split=full/near-term;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":41},{"id":"pub-8a0b1d1ab96f540c070601c9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.424242424242422,"rawValue":0.22424242424242422,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[84];split=full/full-spectrum;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":42},{"id":"pub-88a6dd7250399db905f0ad9f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[134];split=full/last-exam;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":43},{"id":"pub-844cc710f2801277fba3dba0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[306];split=licensed;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":44},{"id":"pub-7c308b545d3aab7b9fa47dfa","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.095238095238095,"rawValue":0.18095238095238095,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[393];split=linux_only;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":45},{"id":"pub-7b75199eee3d3b244d3640d1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[133];split=full/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":46},{"id":"pub-78b769f6172159fa205e9160","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.404255319148938,"rawValue":0.23404255319148937,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[442];split=unlicensed;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":47},{"id":"pub-7357f7d291d91f15bad01556","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":37.3134328358209,"rawValue":0.373134328358209,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[613];split=unlicensed/near-term;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":48},{"id":"pub-72515ce73e869e22f2316772","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[147];split=full/last-exam;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":49},{"id":"pub-6e84ef0f25059c5388926f44","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[324];split=licensed;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":50},{"id":"pub-6bcb1e4a036a026d81576ce1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[328];split=licensed;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":51},{"id":"pub-6b6c4c0b034765d560df4a1f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.723404255319153,"rawValue":0.2872340425531915,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[432];split=unlicensed;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":52},{"id":"pub-67e1614bb36ff1348cbbdaf0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.142857142857142,"rawValue":0.2714285714285714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[366];split=linux_only;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":53},{"id":"pub-67b305a59857aaef1c7564a2","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.644736842105267,"rawValue":0.26644736842105265,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[14];split=full;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":54},{"id":"pub-65209da35e678cf12034290a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.857142857142858,"rawValue":0.22857142857142856,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[382];split=linux_only;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":55},{"id":"pub-61444b9c971a4a6e1d3bfa27","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.0,"rawValue":0.22,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[495];split=unlicensed/full-spectrum;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":56},{"id":"pub-60a9621be6d895a75fa5511a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.850746268656714,"rawValue":0.29850746268656714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[214];split=full/near-term;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":57},{"id":"pub-5ff015662c4b2357dc5e2114","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.723404255319153,"rawValue":0.2872340425531915,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[668];split=unlicensed/overall;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":58},{"id":"pub-5bd60d8acc4afc95b47fd80d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":44.02985074626866,"rawValue":0.44029850746268656,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[603];split=unlicensed/near-term;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":59},{"id":"pub-576f46906e359206750cc663","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.404255319148938,"rawValue":0.23404255319148937,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[678];split=unlicensed/overall;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":60},{"id":"pub-50af39985d7f11e94978adb6","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.850746268656714,"rawValue":0.29850746268656714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[632];split=unlicensed/near-term;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":61},{"id":"pub-491772ed2974f5ba39af7baf","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.0,"rawValue":0.2,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[391];split=linux_only;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":62},{"id":"pub-48eaab608fdfedd7a7589068","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[307];split=licensed;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":63},{"id":"pub-48baba81bf7b892d0fc5bfd9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":39.054726368159194,"rawValue":0.3905472636815919,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[614];split=unlicensed/near-term;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":64},{"id":"pub-47c8f891b40d11a2892386d8","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.026315789473685,"rawValue":0.23026315789473684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[17];split=full;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":65},{"id":"pub-45273783fc2689668b7c3ffa","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.113475177304963,"rawValue":0.24113475177304963,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[435];split=unlicensed;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":66},{"id":"pub-43b67927245ae58c5d3d1b39","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.04964539007092,"rawValue":0.23049645390070922,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[683];split=unlicensed/overall;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":67},{"id":"pub-408d40b3f198dc7f17e5f9fd","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.0,"rawValue":0.18,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[509];split=unlicensed/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":68},{"id":"pub-3f93fe5510cbdee4a826f88f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":33.582089552238806,"rawValue":0.3358208955223881,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[622];split=unlicensed/near-term;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":69},{"id":"pub-398807bbfe33100598f10f8d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.439716312056735,"rawValue":0.18439716312056736,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[693];split=unlicensed/overall;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":70},{"id":"pub-390f188828eba0f9315e527c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.090909090909092,"rawValue":0.09090909090909091,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[310];split=licensed;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":71},{"id":"pub-34cad9e099d658ad716fb75c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.105263157894736,"rawValue":0.17105263157894737,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[275];split=full/overall;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":72},{"id":"pub-347f7ff89b2c571d6e2cd573","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[152];split=full/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":73},{"id":"pub-32d574b7bf1975691dbc1208","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.38157894736842,"rawValue":0.2138157894736842,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[29];split=full;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":74},{"id":"pub-31793bb614cc5fec4e8cdc70","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":16.0,"rawValue":0.16,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[513];split=unlicensed/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":75},{"id":"pub-2add8dfd12b6ce46dad41891","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":33.582089552238806,"rawValue":0.3358208955223881,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[204];split=full/near-term;harness=cursor_cli;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"cursor_cli","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":76},{"id":"pub-2302f983e9feb22906dec84c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[136];split=full/last-exam;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":77},{"id":"pub-1f9143aff81ef0c48bd00089","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.45175438596491,"rawValue":0.2445175438596491,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[256];split=full/overall;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":78},{"id":"pub-1e4dc1a6f37e7c1211382fdf","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[556];split=unlicensed/last-exam;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":79},{"id":"pub-1dcec5d8aa2972aa1f28a335","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.181818181818183,"rawValue":0.18181818181818182,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[93];split=full/full-spectrum;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":80},{"id":"pub-1d8cc0e3bdc8fa388a6bdf6c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.03030303030303,"rawValue":0.0303030303030303,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[309];split=licensed;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":81},{"id":"pub-1c024bb13377496ef8234973","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.858156028368796,"rawValue":0.19858156028368795,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[453];split=unlicensed;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":82},{"id":"pub-17cad275de588ae8de3564cc","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.644736842105267,"rawValue":0.26644736842105265,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[250];split=full/overall;harness=codex;model=gpt-5-5;variant=reasoning-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-xhigh","source_harness":"codex","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":83},{"id":"pub-1700f82939709812f618c036","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":16.363636363636363,"rawValue":0.16363636363636364,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[95];split=full/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-low;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-low","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":84},{"id":"pub-15df3bf3600706c8ecd8f04c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.736842105263158,"rawValue":0.19736842105263158,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[271];split=full/overall;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":85},{"id":"pub-153f3f85f451903a2a4ef2d6","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.421052631578945,"rawValue":0.18421052631578946,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[33];split=full;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":86},{"id":"pub-14d74bcfbe9b3aba2f20a81a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":32.83582089552239,"rawValue":0.3283582089552239,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[194];split=full/near-term;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":87},{"id":"pub-112e725845f3c1125b4a0aae","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.142857142857142,"rawValue":0.17142857142857143,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[392];split=linux_only;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":88},{"id":"pub-10ecf66853acc8e711f1983d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.761904761904763,"rawValue":0.24761904761904763,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[369];split=linux_only;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":89},{"id":"pub-0e75baa14d29025cc475e662","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[555];split=unlicensed/last-exam;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":90},{"id":"pub-0c9d68ae6d96c1325a9c384b","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.636363636363637,"rawValue":0.23636363636363636,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[75];split=full/full-spectrum;harness=ale_claw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"ale_claw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":91},{"id":"pub-080362a9c422096de44d7c8e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":39.054726368159194,"rawValue":0.3905472636815919,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[196];split=full/near-term;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":92},{"id":"pub-0602c7603babdd61b872fffb","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.0,"rawValue":0.18,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[512];split=unlicensed/full-spectrum;harness=openclaw;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":93},{"id":"pub-045496f3e5c1aea7ae123663","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[552];split=unlicensed/last-exam;harness=codex;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"codex","harness_variant":null,"source_harness":"codex","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":94},{"id":"pub-03aa26b01bb0e5442ebaf465","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.181818181818183,"rawValue":0.18181818181818182,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[88];split=full/full-spectrum;harness=droid;model=gpt-5-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"droid","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":95},{"id":"pub-0236de841c1e7f6839fb00f8","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":16.363636363636363,"rawValue":0.16363636363636364,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[94];split=full/full-spectrum;harness=codex;model=gpt-5-5;variant=reasoning-medium;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"codex","harness_variant":"reasoning-medium","source_harness":"codex","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"codex","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":96},{"id":"pub-f25b09d5c3b197a8f084199d","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":51.73999042242261,"rawValue":0.5173999042242261,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[664];split=unlicensed/overall;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ec038418e9f8836e55e0c569","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.58014629586833,"rawValue":0.44580146295868334,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[295];split=licensed;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e830857aa560628d23660dbd","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":71.86616124141027,"rawValue":0.7186616124141026,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[187];split=full/near-term;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-e77eb528b0b87194ff5c1cd4","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":51.22184380800093,"rawValue":0.5122184380800093,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[6];split=full;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-cdd9228b3c2d0d9bf9bb7cb5","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.59063643458926,"rawValue":0.5059063643458925,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[363];split=linux_only;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-aae752e429b9184ec47aabbb","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.56625622278024,"rawValue":0.4456625622278024,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[65];split=full/full-spectrum;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-a5bec53d7e5381a760a949d8","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":51.73999042242261,"rawValue":0.5173999042242261,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[428];split=unlicensed;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-657b5983c40cac1df2402ec4","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":19.67489930133485,"rawValue":0.1967489930133485,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[541];split=unlicensed/last-exam;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-55747b6e64f8fc164d8fe22b","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.700752323542037,"rawValue":0.23700752323542038,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[119];split=full/last-exam;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-4ae1c5c409b3b18b1b5d6dce","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.47048184505826,"rawValue":0.4247048184505826,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[488];split=unlicensed/full-spectrum;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10},{"id":"pub-2174daac414e6d48624f4f6e","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":51.22184380800093,"rawValue":0.5122184380800093,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[242];split=full/overall;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":11},{"id":"pub-0883694ab4b64b561aa04dc7","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":71.86616124141027,"rawValue":0.7186616124141026,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[605];split=unlicensed/near-term;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":12},{"id":"pub-f0d77113d2de9cf754e489fb","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.973684210526315,"rawValue":0.26973684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[6];split=full;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dcf3ed0dc8f6c973e2373ca4","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.0,"rawValue":0.22,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[488];split=unlicensed/full-spectrum;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b5d540cf7f40be4e09f52061","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":38.80597014925373,"rawValue":0.3880597014925373,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[187];split=full/near-term;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-b464be43da3edf29301b68c9","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":38.80597014925373,"rawValue":0.3880597014925373,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[605];split=unlicensed/near-term;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-aa61b483c2fcc78de8561aae","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.0,"rawValue":0.1,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[541];split=unlicensed/last-exam;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-65bd62a9baf4522d41479c36","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.659574468085108,"rawValue":0.2765957446808511,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[664];split=unlicensed/overall;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-53820e84074364c6ba721235","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.636363636363637,"rawValue":0.23636363636363636,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[65];split=full/full-spectrum;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-4c6834b678a837f5caa472a9","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.526315789473683,"rawValue":0.10526315789473684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[119];split=full/last-exam;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-2efc6a326d80e268f6025a30","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":25.71428571428571,"rawValue":0.2571428571428571,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[363];split=linux_only;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-2d5b2bd9b82de8e2066a441c","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.659574468085108,"rawValue":0.2765957446808511,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[428];split=unlicensed;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10},{"id":"pub-15f178a26b31f7742b8af121","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.973684210526315,"rawValue":0.26973684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[242];split=full/overall;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":11},{"id":"pub-0cce4a1491b2d433ae17f837","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":18.181818181818183,"rawValue":0.18181818181818182,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[295];split=licensed;harness=grok_build;model=grok-4-5;variant=reasoning-high;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":"reasoning-high","source_harness":"grok_build","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":12},{"id":"pub-e8e8319929c08f91acdefaa3","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":18.828448362311413,"rawValue":0.18828448362311412,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[530];split=unlicensed/full-spectrum;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e51fd89df2b8580c0ce1f70c","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":1.640722406550101,"rawValue":0.01640722406550101,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[173];split=full/last-exam;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-cee40f467499f39db1d3076d","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":18.20768032937401,"rawValue":0.1820768032937401,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[112];split=full/full-spectrum;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-cd8049275c928980752933e0","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":21.23825941816749,"rawValue":0.2123825941816749,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[412];split=linux_only;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-cbe911c59ff0103cd13e0965","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":22.929688176112588,"rawValue":0.2292968817611259,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[707];split=unlicensed/overall;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-b3f6964a5c62c1ac2f575c47","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":35.12706887636276,"rawValue":0.3512706887636276,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[229];split=full/near-term;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-97855e2f3517c2b9bc8c7d0f","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":21.665039689683386,"rawValue":0.21665039689683385,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[289];split=full/overall;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-72fca6480d7f90a313399d6c","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":2.078248381630128,"rawValue":0.020782483816301277,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[591];split=unlicensed/last-exam;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-6b286d939e104f758b5e0a8d","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":35.12706887636276,"rawValue":0.3512706887636276,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[647];split=unlicensed/near-term;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-4e43b3bd1c978e8d60babfd1","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":5.454545454545454,"rawValue":0.05454545454545454,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[344];split=licensed;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-4c36257f7de1b29a8bfec790","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":22.929688176112588,"rawValue":0.2292968817611259,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[471];split=unlicensed;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-1ab8076d8c8de79c3be3208f","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":21.665039689683386,"rawValue":0.21665039689683385,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[53];split=full;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-c99abe0cdd467065c533bfaf","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":15.671641791044777,"rawValue":0.15671641791044777,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[647];split=unlicensed/near-term;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c576f6d162457339248d6911","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.210526315789473,"rawValue":0.09210526315789473,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[289];split=full/overall;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-c15381d3a086a5289cb3cc95","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[173];split=full/last-exam;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-9d4db1cb4dbad058b1c3c099","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.929078014184398,"rawValue":0.09929078014184398,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[471];split=unlicensed;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-8e227924ff33f5881718031b","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[591];split=unlicensed/last-exam;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-6b674bba8e058655a67d65a3","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":7.000000000000001,"rawValue":0.07,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[530];split=unlicensed/full-spectrum;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-418c399d612cb96c5205890d","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":8.095238095238095,"rawValue":0.08095238095238096,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[412];split=linux_only;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-3adda6de68785bcc8aa9b8f2","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":6.363636363636363,"rawValue":0.06363636363636363,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[112];split=full/full-spectrum;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-1f566c6090a3d9dc3a567b32","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":15.671641791044777,"rawValue":0.15671641791044777,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[229];split=full/near-term;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-1dd0045adbcd3bc038780c4a","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[344];split=licensed;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-14852aa8f327773f92ce78b6","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.210526315789473,"rawValue":0.09210526315789473,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[53];split=full;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-07aa390d356533c9fd4996d9","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2-6","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.929078014184398,"rawValue":0.09929078014184398,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[707];split=unlicensed/overall;harness=openclaw;model=kimi-k2-6;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-fb0a9218c492c80e4a74e664","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":51.583231877805204,"rawValue":0.515832318778052,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[5];split=full;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e4187cb193fa4b698bfdd707","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":71.59801555439591,"rawValue":0.7159801555439591,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[606];split=unlicensed/near-term;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-e2903a242fe069a68fed0d59","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":20.49303255658983,"rawValue":0.2049303255658983,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[122];split=full/last-exam;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-cd54bd1303748c364b4ac0f7","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.78637305133695,"rawValue":0.5278637305133695,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[356];split=linux_only;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-c442c47a1eb3e0889ded42e6","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":69.68586979044264,"rawValue":0.6968586979044264,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[609];split=unlicensed/near-term;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-bd5b0a4977c5e9963d1dc380","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.72487812208666,"rawValue":0.5072487812208666,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[7];split=full;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-a9aec0c39db4ccc4c9f91cd8","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":20.075451881442678,"rawValue":0.2007545188144268,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[540];split=unlicensed/last-exam;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-a265ea2b99bc8c88db783a36","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.69304388437668,"rawValue":0.5269304388437668,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[661];split=unlicensed/overall;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-a19e2231105b5be7e32da2d4","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.69304388437668,"rawValue":0.5269304388437668,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[425];split=unlicensed;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-91c92cb16cdd4b1896eff549","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":32.97649487170496,"rawValue":0.3297649487170496,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[296];split=licensed;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-91590fcf22c04a966db5d96b","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":25.496571532732816,"rawValue":0.25496571532732815,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[301];split=licensed;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-73e6d5156fda239b91a640b9","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":51.583231877805204,"rawValue":0.515832318778052,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[241];split=full/overall;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-6923c7c434176ab95ce220d8","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":48.133859388693956,"rawValue":0.48133859388693956,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[477];split=unlicensed/full-spectrum;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-60aa253e2a616f881e6a7983","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":51.95649529905552,"rawValue":0.5195649529905552,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[358];split=linux_only;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-5ac4cf39e7ee5b0e543a16af","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.940530822949675,"rawValue":0.44940530822949676,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[64];split=full/full-spectrum;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-54fa09b11268dd8c2b2fd91d","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":20.55276315873776,"rawValue":0.2055276315873776,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[121];split=full/last-exam;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-4884f8c5e7eb25cdcc397e18","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":49.87405398972178,"rawValue":0.4987405398972178,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[59];split=full/full-spectrum;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-42df3c93f19d2991a2738419","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":50.72487812208666,"rawValue":0.5072487812208666,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[243];split=full/overall;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-3725ad995e818a84a65c7794","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.142431676345083,"rawValue":0.23142431676345085,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[538];split=unlicensed/last-exam;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-2d32f66455e4a2f0505c7b76","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":53.03482128962862,"rawValue":0.5303482128962862,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[424];split=unlicensed;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-1f03a0e3d98a8cca4494dca9","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.51458390524464,"rawValue":0.4351458390524464,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[485];split=unlicensed/full-spectrum;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-19f4b38fde6c4b835cf328d8","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":71.59801555439591,"rawValue":0.7159801555439591,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[188];split=full/near-term;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-0bcadc399c9aa05e72d39e27","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":69.68586979044264,"rawValue":0.6968586979044264,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[191];split=full/near-term;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-08dd0123bb050166f4f4bc6d","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":53.03482128962862,"rawValue":0.5303482128962862,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[660];split=unlicensed/overall;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-f852d65dfd2704c4b66b526a","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.09090909090909,"rawValue":0.2909090909090909,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[59];split=full/full-spectrum;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f663dddfaf9cd1a5b6827234","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.61904761904762,"rawValue":0.2761904761904762,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[356];split=linux_only;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-f65517744077ad7aa69c26c8","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.0,"rawValue":0.1,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[538];split=unlicensed/last-exam;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-e9ebdc8fad2b98569db20653","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.368794326241137,"rawValue":0.28368794326241137,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[661];split=unlicensed/overall;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-e882ec4a0551d8c919f20176","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.526315789473683,"rawValue":0.10526315789473684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[121];split=full/last-exam;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-dde400cbed85efcfa88b312b","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.368794326241137,"rawValue":0.28368794326241137,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[425];split=unlicensed;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-c57a99ba4ea98784b5fd2d30","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":40.298507462686565,"rawValue":0.40298507462686567,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[188];split=full/near-term;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-b70e301d425fab40c859bf2b","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.289473684210524,"rawValue":0.28289473684210525,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[241];split=full/overall;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8},{"id":"pub-b4ad6c25991c111be8c53161","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.333333333333334,"rawValue":0.13333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[540];split=unlicensed/last-exam;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":9},{"id":"pub-8e78e28d908b4c43c17d79b6","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.78723404255319,"rawValue":0.2978723404255319,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[660];split=unlicensed/overall;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10},{"id":"pub-8df5ea7eb4855dbef91b6c67","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":38.80597014925373,"rawValue":0.3880597014925373,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[609];split=unlicensed/near-term;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11},{"id":"pub-7bb5759e62daf480e4721714","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.78723404255319,"rawValue":0.2978723404255319,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[424];split=unlicensed;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12},{"id":"pub-7a70478c3fa720a14a351f04","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":7.894736842105263,"rawValue":0.07894736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[122];split=full/last-exam;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13},{"id":"pub-65c3e1453c6668ba43a69cc8","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.000000000000004,"rawValue":0.28,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[477];split=unlicensed/full-spectrum;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":14},{"id":"pub-59eaf3f1fea483fb0ded9696","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.636363636363637,"rawValue":0.23636363636363636,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[64];split=full/full-spectrum;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":15},{"id":"pub-5837a6e88b8b4a47f704a459","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.090909090909092,"rawValue":0.09090909090909091,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[296];split=licensed;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":16},{"id":"pub-554de4d627ed1c3aeda9f6b7","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":27.61904761904762,"rawValue":0.2761904761904762,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[358];split=linux_only;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":17},{"id":"pub-4c0179ad9ca11bbc91ef708c","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":38.80597014925373,"rawValue":0.3880597014925373,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[191];split=full/near-term;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":18},{"id":"pub-4638d257d7d5a26f9bec0e2d","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":28.289473684210524,"rawValue":0.28289473684210525,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[5];split=full;harness=kimi_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"kimi-code","harness_variant":"thinking-max","source_harness":"kimi_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"kimi-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":19},{"id":"pub-456526ad7d409815ce604307","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.973684210526315,"rawValue":0.26973684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[243];split=full/overall;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":20},{"id":"pub-3c29fb9fe05ab1bd76803961","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.973684210526315,"rawValue":0.26973684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[7];split=full;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":21},{"id":"pub-2e7938d6cc89333e1c475b5a","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":40.298507462686565,"rawValue":0.40298507462686567,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[606];split=unlicensed/near-term;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":22},{"id":"pub-288bc7c091c7730f9195c161","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.090909090909092,"rawValue":0.09090909090909091,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[301];split=licensed;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":23},{"id":"pub-1b38e767bc7c18c65ed8ceb3","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.0,"rawValue":0.22,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[485];split=unlicensed/full-spectrum;harness=claude_code;model=kimi-k3;variant=thinking-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-max","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":24},{"id":"pub-e5314f14194590dff4f969b2","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":21.744599999999995,"rawValue":0.21744599999999994,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[528];split=unlicensed/full-spectrum;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d644e545a04ba2c50824bd53","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.555956574853624,"rawValue":0.23555956574853623,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[288];split=full/overall;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d0db3037301306f20b9590b8","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":26.465217947397846,"rawValue":0.26465217947397845,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[410];split=linux_only;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-cac4f2f20d998f6d766fda00","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":23.555956574853624,"rawValue":0.23555956574853623,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[52];split=full;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-bce9ad0f54243ce2ecc113df","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":35.08749428062655,"rawValue":0.3508749428062655,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[648];split=unlicensed/near-term;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-a8eb1a95304a12d16b893042","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":20.754532410468567,"rawValue":0.20754532410468568,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[110];split=full/full-spectrum;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-82013fe23c1135f344570aa7","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":25.00876678582964,"rawValue":0.2500876678582964,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[706];split=unlicensed/overall;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-7b33cb2365a9e8a153fe66b9","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":25.00876678582964,"rawValue":0.2500876678582964,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[470];split=unlicensed;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-6ab1bcc2940cc1d55a1d5749","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":4.9335711432519425,"rawValue":0.04933571143251942,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[345];split=licensed;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-5896b8d3071b45d0952d8395","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":35.08749428062655,"rawValue":0.3508749428062655,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[230];split=full/near-term;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-335d7931794c469ad0e73a46","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":4.3168,"rawValue":0.043168,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[584];split=unlicensed/last-exam;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-0743421a9f6679f3f8cb065b","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":3.408,"rawValue":0.03408,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[167];split=full/last-exam;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-fe767db30216600765f74a76","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":8.552631578947368,"rawValue":0.08552631578947369,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[288];split=full/overall;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e615a8ce7160d4cca31e377e","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.219858156028367,"rawValue":0.09219858156028368,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[706];split=unlicensed/overall;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-c16fef91bc306edc131f2840","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[167];split=full/last-exam;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-a96e5be366632d0a05effe10","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.090909090909092,"rawValue":0.09090909090909091,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[110];split=full/full-spectrum;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-9009df64116cc2eadd50e10f","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.219858156028367,"rawValue":0.09219858156028368,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[470];split=unlicensed;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-8f93c4187d8ef03d90f3baea","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.0,"rawValue":0.1,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[410];split=linux_only;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-8d251286ad9d76bb4832f2c1","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":8.552631578947368,"rawValue":0.08552631578947369,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[52];split=full;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-6085b66d68eaa719aa9ad7e6","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[345];split=licensed;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-55834ba24b1e0c435ec08605","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.940298507462686,"rawValue":0.11940298507462686,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[648];split=unlicensed/near-term;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-32084d3584bc933f585dba06","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.940298507462686,"rawValue":0.11940298507462686,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[230];split=full/near-term;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-279119b4f8a0fe35f3c7186b","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.0,"rawValue":0.1,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[528];split=unlicensed/full-spectrum;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-1215e6598f90c7815bb72027","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2-5","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[584];split=unlicensed/last-exam;harness=openclaw;model=mimo-v2-5;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-f0cbbd9fe3f9fc4c46e4d5a9","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":1.6203333333333334,"rawValue":0.016203333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[592];split=unlicensed/last-exam;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-dce56a2d22221ce6ca960abd","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":24.547984743567735,"rawValue":0.24547984743567736,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[650];split=unlicensed/near-term;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d2755e31fd8a24b80a78a03f","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.92606367247545,"rawValue":0.1492606367247545,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[474];split=unlicensed;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-a347974681c81045fd90345e","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":1.2792105263157894,"rawValue":0.012792105263157895,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[174];split=full/last-exam;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-7fea35bebb3ed684d04b5925","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.64903265101336,"rawValue":0.1464903265101336,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[416];split=linux_only;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-6f5dc2dd04061f25bc5b0129","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.187993275125251,"rawValue":0.14187993275125252,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[292];split=full/overall;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-6e98581ea7490b6ac27c6613","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.225000000000001,"rawValue":0.08225000000000002,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[534];split=unlicensed/full-spectrum;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-6188adf374b274b0db659d8e","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.422727272727274,"rawValue":0.08422727272727273,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[116];split=full/full-spectrum;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-51cc0ce91c449bb46a93f031","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":24.547984743567735,"rawValue":0.24547984743567736,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[232];split=full/near-term;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-4ae705a0163bb00669aaadae","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":4.7272727272727275,"rawValue":0.04727272727272727,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[346];split=licensed;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-3f0398941161bc6aa6b6e9d0","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.92606367247545,"rawValue":0.1492606367247545,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[710];split=unlicensed/overall;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-34bb7cd010a3aba12f772841","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.187993275125251,"rawValue":0.14187993275125252,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[56];split=full;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-dce7b6bbc86087f8fe731608","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[592];split=unlicensed/last-exam;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b26b43ed50ea06b3429cc82f","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":5.921052631578947,"rawValue":0.05921052631578947,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[56];split=full;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-afcab69be4539d6de9a9d23a","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.44776119402985,"rawValue":0.1044776119402985,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[232];split=full/near-term;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-aa46d4e0866a6ec53b2673f1","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":5.714285714285714,"rawValue":0.05714285714285714,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[416];split=linux_only;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-9e6388595549f1487f6184ed","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[346];split=licensed;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-913cde44d362651a3ba38d8e","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":4.0,"rawValue":0.04,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[534];split=unlicensed/full-spectrum;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-8ad91645b23ea4c61e7e1064","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[174];split=full/last-exam;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-86009f935aa8d4076be80a18","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":6.382978723404255,"rawValue":0.06382978723404255,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[474];split=unlicensed;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-7df899d18ee17c6b998b3ef2","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.6363636363636362,"rawValue":0.03636363636363636,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[116];split=full/full-spectrum;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-7985e5a89b9af2cc6c8f0336","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":5.921052631578947,"rawValue":0.05921052631578947,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[292];split=full/overall;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-5fc65d84813a2ff1b0a93134","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":6.382978723404255,"rawValue":0.06382978723404255,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[710];split=unlicensed/overall;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-2d74ff1d327f419326f488d4","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2-7","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.44776119402985,"rawValue":0.1044776119402985,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[650];split=unlicensed/near-term;harness=openclaw;model=minimax-m2-7;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-f428766438793db87c5c47c2","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.54068132927283,"rawValue":0.4354068132927283,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[381];split=linux_only;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e8cf25002efedec4450a5c8b","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":14.726424569791586,"rawValue":0.14726424569791585,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[131];split=full/last-exam;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-da1a3150214aa5ebded8ccc0","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":13.893693343958233,"rawValue":0.13893693343958233,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[551];split=unlicensed/last-exam;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-d2f2016a94549d0ba870c874","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":60.02555857554045,"rawValue":0.6002555857554045,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[630];split=unlicensed/near-term;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-b81e78692fde370d4e06601d","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.59353752236476,"rawValue":0.4459353752236476,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[440];split=unlicensed;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-b488cf92e288ee999560d6f3","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":44.59353752236476,"rawValue":0.4459353752236476,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[676];split=unlicensed/overall;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-adc96de822791ee08490bce9","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.8847508157024,"rawValue":0.428847508157024,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[257];split=full/overall;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-66eaecda79f6df347ddf4b3d","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.8847508157024,"rawValue":0.428847508157024,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[21];split=full;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-541a42c369df819bcfcc2767","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":43.039726926162565,"rawValue":0.43039726926162564,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[487];split=unlicensed/full-spectrum;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-3c4b438fb13b1dc33fc88fa5","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":60.02555857554045,"rawValue":0.6002555857554045,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[212];split=full/near-term;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-22da0551701e0ac0cfcd48d3","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":42.54520629651143,"rawValue":0.4254520629651143,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[70];split=full/full-spectrum;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-08d1a3c1d28245166029a134","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":20.981212121212124,"rawValue":0.20981212121212123,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[312];split=licensed;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-eb98d81d07aa8f8957bc44ed","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":31.343283582089555,"rawValue":0.31343283582089554,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[630];split=unlicensed/near-term;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-db99a36538cf40dcb85e252e","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[312];split=licensed;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d27be26eedf3b5cf1a8bdce3","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":22.0,"rawValue":0.22,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[487];split=unlicensed/full-spectrum;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-cd3bd68a3a538ed1a5ee8599","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":3.3333333333333335,"rawValue":0.03333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[551];split=unlicensed/last-exam;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-adbf21ba10ee7b24cceee447","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":31.343283582089555,"rawValue":0.31343283582089554,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[212];split=full/near-term;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-aab6f38697649d96f98c4768","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.98581560283688,"rawValue":0.2198581560283688,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[676];split=unlicensed/overall;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-9803ba36ebe45f006d7fc6d7","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.818181818181817,"rawValue":0.21818181818181817,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[70];split=full/full-spectrum;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-7bfe1263c5850ff90effdddc","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":2.631578947368421,"rawValue":0.02631578947368421,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[131];split=full/last-exam;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-603b44dc756f66a2588b7f35","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":19.047619047619047,"rawValue":0.19047619047619047,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[381];split=linux_only;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-55b6225a7777ddab96c3b232","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.394736842105264,"rawValue":0.20394736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[21];split=full;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-13e19d2a80f85127cc568923","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":20.394736842105264,"rawValue":0.20394736842105263,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[257];split=full/overall;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-0b0868699b36975af33e4514","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen-3-8-27b","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":21.98581560283688,"rawValue":0.2198581560283688,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[440];split=unlicensed;harness=claude_code;model=qwen-3-8-27b;variant=reasoning-max;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":null,"evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"reasoning-max","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-f8e4c4942c68ebeb56ac07c0","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":55.260263815045064,"rawValue":0.5526026381504506,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[655];split=unlicensed/overall;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d2ddbe8f0f6820cdad76f66a","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.354442255769975,"rawValue":0.46354442255769973,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[61];split=full/full-spectrum;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-aba30c1dc747f8011febe368","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.48561340297907,"rawValue":0.5248561340297907,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[238];split=full/overall;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-917614acbc3f502e032a5b0d","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.727232361384374,"rawValue":0.46727232361384374,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[478];split=unlicensed/full-spectrum;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-858689101d1f123263f1bd4b","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":72.1190499337615,"rawValue":0.721190499337615,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[604];split=unlicensed/near-term;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-7e357924fa2124b8d2c980bd","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":24.1291868854736,"rawValue":0.241291868854736,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[118];split=full/last-exam;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-726b505a0a230fe8f3ab309a","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":52.48561340297907,"rawValue":0.5248561340297907,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[2];split=full;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-57a444f2b20e7d8258b3de23","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":72.1190499337615,"rawValue":0.721190499337615,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[186];split=full/near-term;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-2786935ffa2a143b7451d10a","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":16.919639939223963,"rawValue":0.16919639939223963,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[322];split=licensed;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-1d514ce716421dba4ed8df81","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":28.13085894382211,"rawValue":0.28130858943822107,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[536];split=unlicensed/last-exam;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-13f70f68525e987dc3f9f5e1","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":55.260263815045064,"rawValue":0.5526026381504506,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[419];split=unlicensed;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-0588d14ffb9c1eb5800e86cd","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":51.12101968778015,"rawValue":0.5112101968778014,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[360];split=linux_only;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-ce445d401a68db0d5300aec4","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.078014184397162,"rawValue":0.2907801418439716,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[655];split=unlicensed/overall;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c71929a5b39f7c2e03412568","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.973684210526315,"rawValue":0.26973684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[238];split=full/overall;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-c45231103580b62a6aa1f2e9","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":38.80597014925373,"rawValue":0.3880597014925373,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[604];split=unlicensed/near-term;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b4b47586ba6276e049e54bc0","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":23.636363636363637,"rawValue":0.23636363636363636,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[61];split=full/full-spectrum;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-98e4700a8c50f9ea7f2bded0","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":26.973684210526315,"rawValue":0.26973684210526316,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[2];split=full;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-989dc70f1480a5d047ad67f5","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[322];split=licensed;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-41cbd96351d110d9513565d1","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.0,"rawValue":0.24,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[478];split=unlicensed/full-spectrum;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-2a295c4ccc1b71a4f4129c45","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":24.761904761904763,"rawValue":0.24761904761904763,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[360];split=linux_only;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-285b3b0c506c72ce6f01c93e","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.333333333333334,"rawValue":0.13333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[536];split=unlicensed/last-exam;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9},{"id":"pub-21c6cead33aa2af365038f15","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":29.078014184397162,"rawValue":0.2907801418439716,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[419];split=unlicensed;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-064e6e0ecae307f9da621cd4","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.526315789473683,"rawValue":0.10526315789473684,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[118];split=full/last-exam;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":11},{"id":"pub-00da1a25423a94340d8cba5e","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen-3-8-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":38.80597014925373,"rawValue":0.3880597014925373,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[186];split=full/near-term;harness=claude_code;model=qwen-3-8-max;variant=thinking-xhigh;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"claude-code","harness_variant":"thinking-xhigh","source_harness":"claude_code","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["agent_system_score","harness_specific","heuristic_model_mapping","missing_source_date","split_specific"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":12},{"id":"pub-ef35a8b148e44b6a88a06eff","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":30.380852488057243,"rawValue":0.3038085248805724,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[515];split=unlicensed/full-spectrum;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c6600bdfb54d16080b6524f7","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.89539000915916,"rawValue":0.46895390009159155,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[224];split=full/near-term;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c5d89e6fe75a21b3a1416df1","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.691021666783875,"rawValue":0.33691021666783877,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[697];split=unlicensed/overall;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-b5f57d56beb206f05cc9534e","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":28.491684080052032,"rawValue":0.2849168408005203,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[100];split=full/full-spectrum;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-9c56c39fc1c43c2a7449bf60","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":34.63812505854651,"rawValue":0.34638125058546504,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[398];split=linux_only;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-9c2311cf43d0d3869fdf8fa9","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":31.568645098792942,"rawValue":0.3156864509879294,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[281];split=full/overall;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-862ba707b7e9473167773ac1","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":33.691021666783875,"rawValue":0.33691021666783877,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[461];split=unlicensed;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-6320bfc742dd7ee25123a2e0","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":6.390764530880215,"rawValue":0.06390764530880215,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[161];split=full/last-exam;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-364c44e9b8f56bb6e6c9d6b3","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":31.568645098792942,"rawValue":0.3156864509879294,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[45];split=full;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-1bf3a7f7f5905fb899478b3e","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":4.363636363636363,"rawValue":0.04363636363636363,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[347];split=licensed;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10},{"id":"pub-0d4c9fc6ca0891fd6935ef1e","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":46.89539000915916,"rawValue":0.46895390009159155,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[642];split=unlicensed/near-term;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":11},{"id":"pub-0bab71f706949de7a145f5b7","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":8.094968405781607,"rawValue":0.08094968405781607,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[577];split=unlicensed/last-exam;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":12},{"id":"pub-e35ea090a2fe30f23a25fca5","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":13.333333333333334,"rawValue":0.13333333333333333,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[398];split=linux_only;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cba7c708dd10cb5ae6c36463","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[161];split=full/last-exam;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-bcf905e6f892904ae60d1f19","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[577];split=unlicensed/last-exam;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-aca635ee1d6d752cbf6e7b48","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.842105263157894,"rawValue":0.11842105263157894,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[45];split=full;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-8c9f41edb00afb8498dccad0","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.76595744680851,"rawValue":0.1276595744680851,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[697];split=unlicensed/overall;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-7baaed58e866c9c766d9b134","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":11.842105263157894,"rawValue":0.11842105263157894,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[281];split=full/overall;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-6d8cf11c2f50d09190c0e21b","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.76595744680851,"rawValue":0.1276595744680851,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[461];split=unlicensed;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-571f562ea42434821e9f1a7a","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.0,"rawValue":0.12,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[515];split=unlicensed/full-spectrum;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-3efc3f6f42d3bbbefdf1aa55","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.91044776119403,"rawValue":0.1791044776119403,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[224];split=full/near-term;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-396f4ab7cf6946f2ad6ab4d6","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":17.91044776119403,"rawValue":0.1791044776119403,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[642];split=unlicensed/near-term;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10},{"id":"pub-24710b71792065fbc6a83b78","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.909090909090908,"rawValue":0.10909090909090909,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[100];split=full/full-spectrum;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":11},{"id":"pub-1267f5558705fbdc526b6adc","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen-qwen3-7-max","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[347];split=licensed;harness=openclaw;model=qwen-qwen3-7-max;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db+list","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":12},{"id":"pub-f6767105d6b9b794e702fee2","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":26.021994074929346,"rawValue":0.26021994074929344,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[705];split=unlicensed/overall;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b85bfce593ecc0e6cfbce742","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":6.469090909090909,"rawValue":0.06469090909090909,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[339];split=licensed;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-af2119930547288707824046","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":26.021994074929346,"rawValue":0.26021994074929344,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[469];split=unlicensed;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ab5cf659a72531bf2d122930","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":28.958436379295037,"rawValue":0.2895843637929504,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[409];split=linux_only;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-a9c43120c5f0c00370db8db5","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":22.90770537090957,"rawValue":0.2290770537090957,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[108];split=full/full-spectrum;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-78980a33acc55dc5bafb2ccf","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":36.42233610694047,"rawValue":0.36422336106940467,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[228];split=full/near-term;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-6637535f1ea23fd95083e30e","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":24.606981345822614,"rawValue":0.24606981345822615,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[51];split=full;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-4bd6d899b27afa77a6a46d69","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":4.64254429111199,"rawValue":0.0464254429111199,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[583];split=unlicensed/last-exam;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-45dd52ab47322da58dc547a3","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":36.42233610694047,"rawValue":0.36422336106940467,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[646];split=unlicensed/near-term;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-2f7910b15f1ad5068ab27035","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":24.758475908000527,"rawValue":0.24758475908000527,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[523];split=unlicensed/full-spectrum;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10},{"id":"pub-1edbb3b9b7efed208ad28d7f","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":4.958850756141044,"rawValue":0.049588507561410436,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[165];split=full/last-exam;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":11},{"id":"pub-07ed112c4808f34f7960a317","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"avg_score","value":24.606981345822614,"rawValue":0.24606981345822615,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[287];split=full/overall;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=avg_score","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":12},{"id":"pub-ecb5d7bdd21f4f243825c784","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":8.552631578947368,"rawValue":0.08552631578947369,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[287];split=full/overall;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/overall","split_benchmark_ref":"agents-last-exam-v1-full-overall","split_track":"full/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-db083475472d6c119d877ecf","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":8.181818181818182,"rawValue":0.08181818181818182,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[108];split=full/full-spectrum;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-full-full-spectrum","split_track":"full/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d812c68514c1aaacf7cbfd42","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.219858156028367,"rawValue":0.09219858156028368,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[705];split=unlicensed/overall;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/overall","split_benchmark_ref":"agents-last-exam-v1-unlicensed-overall","split_track":"unlicensed/overall","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-b811205a055192ffb17be886","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[165];split=full/last-exam;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/last-exam","split_benchmark_ref":"agents-last-exam-v1-full-last-exam","split_track":"full/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-b7f9e814a9267c60d5f0c917","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.686567164179104,"rawValue":0.12686567164179105,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[228];split=full/near-term;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full/near-term","split_benchmark_ref":"agents-last-exam-v1-full-near-term","split_track":"full/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-83c9939c9cbe6695c45042a7","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[583];split=unlicensed/last-exam;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/last-exam","split_benchmark_ref":"agents-last-exam-v1-unlicensed-last-exam","split_track":"unlicensed/last-exam","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-642b38a482b23a058bf45c0e","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.0,"rawValue":0.09,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[523];split=unlicensed/full-spectrum;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/full-spectrum","split_benchmark_ref":"agents-last-exam-v1-unlicensed-full-spectrum","split_track":"unlicensed/full-spectrum","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-5e3d409d8c25b727f1b9106b","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":8.552631578947368,"rawValue":0.08552631578947369,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[51];split=full;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"full","split_benchmark_ref":"agents-last-exam-v1-full","split_track":"full","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-39d39fbee2144380cf676de4","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":10.476190476190476,"rawValue":0.10476190476190476,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[409];split=linux_only;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"linux_only","split_benchmark_ref":"agents-last-exam-v1-linux-only","split_track":"linux_only","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-2dde3e111e91f09888a164f2","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":12.686567164179104,"rawValue":0.12686567164179105,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[646];split=unlicensed/near-term;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed/near-term","split_benchmark_ref":"agents-last-exam-v1-unlicensed-near-term","split_track":"unlicensed/near-term","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10},{"id":"pub-2cef481fefca6c988c68e02a","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":0.0,"rawValue":0,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[339];split=licensed;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"licensed","split_benchmark_ref":"agents-last-exam-v1-licensed","split_track":"licensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":11},{"id":"pub-0d7de1fc32c0eba8f706afd0","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3-6-plus","benchmarkId":"agents-last-exam","benchmarkName":"Agents' Last Exam","benchmarkVersion":"ALE-v1","benchmarkVersionId":"agents-last-exam@v1","metricId":"pass_rate","value":9.219858156028367,"rawValue":0.09219858156028368,"unit":"percent","evidenceLevel":"A","sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourcePageUrl":"https://agents-last-exam.org/leaderboard","sourceApiUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[469];split=unlicensed;harness=openclaw;model=qwen3-6-plus;variant=unknown;metric=pass_rate","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","protocol":{"benchmark_version":"ALE-v1","benchmark_version_id":"agents-last-exam@v1","cost_source":"db","evaluator":"agents-last-exam","harness":"agents-last-exam","harness_variant":null,"source_harness":"openclaw","split":"unlicensed","split_benchmark_ref":"agents-last-exam-v1-unlicensed","split_track":"unlicensed","subject_type":"system"},"harnessId":"agents-last-exam","qualityFlags":["agent_system_score","curated_model_alias","harness_specific","missing_harness_variant","missing_source_date","split_specific","unregistered_source_harness"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":12},{"id":"pub-46cca02714c1621ca8b1ad88","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek R1 (0528)","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersionHint":"aider-polyglot@public","metricId":"pass_rate_1","value":34.4,"rawValue":34.4,"unit":"percent","evidenceLevel":"A","sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourcePageUrl":"https://aider.chat/docs/leaderboards/","sourceApiUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1485;dirname=2025-06-06-16-47-07--r1-diff","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-06-06","payloadSha256":"85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de","protocol":{"command":"aider --model deepseek/deepseek-reasoner","edit_format":"diff","harness":"aider","subject_type":"system","test_cases":224},"harnessId":"aider","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-425d6c2399ee46cc8bb6775f","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersion":"2025-05-28","benchmarkVersionHint":"aider-polyglot@public","metricId":"Percent correct","value":71.4,"rawValue":"71.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=35;column=Percent correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-06","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard"},{"id":"pub-8d219c0c0c28d1ff3735856d","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"aider-polyglot@public","metricId":"Percent correct","value":70.2,"rawValue":"70.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=5;column=Percent correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard"},{"id":"pub-169fd93046d81f5a3c59baff","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp (Chat)","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersionHint":"aider-polyglot@public","metricId":"pass_rate_1","value":38.7,"rawValue":38.7,"unit":"percent","evidenceLevel":"A","sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourcePageUrl":"https://aider.chat/docs/leaderboards/","sourceApiUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1830;dirname=2025-10-03-09-21-36--deepseek-v3.2-chat","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-10-03","payloadSha256":"85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de","protocol":{"command":"aider --model deepseek/deepseek-chat","edit_format":"diff","harness":"aider","subject_type":"system","test_cases":225},"harnessId":"aider","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-a4eb189379d402d37b68fb7a","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp (Reasoner)","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersionHint":"aider-polyglot@public","metricId":"pass_rate_1","value":39.6,"rawValue":39.6,"unit":"percent","evidenceLevel":"A","sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourcePageUrl":"https://aider.chat/docs/leaderboards/","sourceApiUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1802;dirname=2025-10-03-09-45-34--deepseek-v3.2-reasoner","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-10-03","payloadSha256":"85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de","protocol":{"command":"aider --model deepseek/deepseek-reasoner","edit_format":"diff","harness":"aider","subject_type":"system","test_cases":225},"harnessId":"aider","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-789c924ce7c57319cc1ea595","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp_thinking","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"aider-polyglot@public","metricId":"Percent correct","value":74.2,"rawValue":"74.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=6;column=Percent correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard"},{"id":"pub-3743508f7334322c858a2830","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"aider-polyglot@public","metricId":"Percent correct","value":70.7,"rawValue":"70.7","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=23;column=Percent correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard"},{"id":"pub-747d737071598c375f75ae9b","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514 (32k thinking)","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersionHint":"aider-polyglot@public","metricId":"pass_rate_1","value":37.3,"rawValue":37.3,"unit":"percent","evidenceLevel":"A","sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourcePageUrl":"https://aider.chat/docs/leaderboards/","sourceApiUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1341;dirname=2025-05-25-20-40-51--opus4-diff-exuser","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-05-25","payloadSha256":"85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de","protocol":{"command":"aider --model claude-opus-4-20250514","edit_format":"diff","harness":"aider","subject_type":"system","test_cases":225},"harnessId":"aider","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-41b74ea27a48b9872097532d","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514 (no think)","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersionHint":"aider-polyglot@public","metricId":"pass_rate_1","value":32.9,"rawValue":32.9,"unit":"percent","evidenceLevel":"A","sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourcePageUrl":"https://aider.chat/docs/leaderboards/","sourceApiUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1313;dirname=2025-05-25-19-57-20--opus4-diff-exuser","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-05-25","payloadSha256":"85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de","protocol":{"command":"aider --model claude-opus-4-20250514","edit_format":"diff","harness":"aider","subject_type":"system","test_cases":225},"harnessId":"aider","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-da0d87078a1863e07665bf8e","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_32K","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"aider-polyglot@public","metricId":"Percent correct","value":72.0,"rawValue":"72.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=24;column=Percent correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard"},{"id":"pub-a8e26eed3c55f8be3257e67a","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-chat","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersion":"2025-12-01","benchmarkVersionHint":"aider-polyglot@public","metricId":"Percent correct","value":70.2,"rawValue":"70.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=78;column=Percent correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-af9d3ba6e98455ec5234fb01","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5 (high)","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersionHint":"aider-polyglot@public","metricId":"pass_rate_1","value":52.0,"rawValue":52.0,"unit":"percent","evidenceLevel":"A","sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourcePageUrl":"https://aider.chat/docs/leaderboards/","sourceApiUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1715;dirname=2025-08-23-15-47-21--gpt-5-high","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-08-23","payloadSha256":"85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de","protocol":{"command":"aider --model openai/gpt-5","edit_format":"diff","harness":"aider","subject_type":"system","test_cases":225},"harnessId":"aider","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9581db8370c1cd126897f14e","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5 (low)","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersionHint":"aider-polyglot@public","metricId":"pass_rate_1","value":43.1,"rawValue":43.1,"unit":"percent","evidenceLevel":"A","sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourcePageUrl":"https://aider.chat/docs/leaderboards/","sourceApiUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1773;dirname=2025-08-25-14-16-37--gpt-5-low","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-08-25","payloadSha256":"85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de","protocol":{"command":"aider --model openai/gpt-5","edit_format":"diff","harness":"aider","subject_type":"system","test_cases":225},"harnessId":"aider","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-8177701022e27455099ba475","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5 (medium)","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersionHint":"aider-polyglot@public","metricId":"pass_rate_1","value":49.8,"rawValue":49.8,"unit":"percent","evidenceLevel":"A","sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourcePageUrl":"https://aider.chat/docs/leaderboards/","sourceApiUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1744;dirname=2025-08-25-13-23-27--gpt-5-medium","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-08-25","payloadSha256":"85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de","protocol":{"command":"aider --model openai/gpt-5","edit_format":"diff","harness":"aider","subject_type":"system","test_cases":225},"harnessId":"aider","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b245a84af2c8e2962cd551d7","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"aider-polyglot@public","metricId":"Percent correct","value":88.0,"rawValue":"88.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=9;column=Percent correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard"},{"id":"pub-92f31d48adc5718581501c1e","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_low","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"aider-polyglot@public","metricId":"Percent correct","value":81.3,"rawValue":"81.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=7;column=Percent correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard"},{"id":"pub-064247470914160460547c12","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"aider-polyglot@public","metricId":"Percent correct","value":86.7,"rawValue":"86.7","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=8;column=Percent correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard"},{"id":"pub-1f67c794b8684edc8813b039","canonicalModelId":"xai/grok-4@current","modelName":"Grok 4","modelRef":"grok-4 (high)","benchmarkId":"aider-polyglot","benchmarkName":"Aider Polyglot","benchmarkVersionHint":"aider-polyglot@public","metricId":"pass_rate_1","value":40.9,"rawValue":40.9,"unit":"percent","evidenceLevel":"A","sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourcePageUrl":"https://aider.chat/docs/leaderboards/","sourceApiUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1630;dirname=2025-07-11-19-37-40--xai-or-grok4-high","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-07-11","payloadSha256":"85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de","protocol":{"command":"aider --model openrouter/x-ai/grok-4","edit_format":"diff","harness":"aider","subject_type":"system","test_cases":225},"harnessId":"aider","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2850b63e237adb7151156e20","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"Qwen/Qwen3.5-27B","benchmarkId":"aime-2026","benchmarkName":"AIME 2026","benchmarkVersionHint":"aime-2026@v1","metricId":"score","value":90.83,"rawValue":90.83,"unit":"percent","evidenceLevel":"C","sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourceLocator":"rank=19;row=18","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"bd8f3e47321d266bfd4c817736e5f0d18cbcd5500ebef2c3b98cffdf31777d0c","protocol":{"filename":".eval_results/MathArena--aime_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://matharena.ai/?comp=aime--aime_2026"},{"id":"pub-d410ddf9441c9cee04309c83","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"Qwen/Qwen3.5-397B-A17B","benchmarkId":"aime-2026","benchmarkName":"AIME 2026","benchmarkVersionHint":"aime-2026@v1","metricId":"score","value":93.33,"rawValue":93.33,"unit":"percent","evidenceLevel":"C","sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourceLocator":"rank=14;row=13","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"bd8f3e47321d266bfd4c817736e5f0d18cbcd5500ebef2c3b98cffdf31777d0c","protocol":{"filename":".eval_results/MathArena--aime_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://matharena.ai/?comp=aime--aime_2026"},{"id":"pub-388ec68b9de326d0244cf6bb","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"Qwen/Qwen3.6-27B","benchmarkId":"aime-2026","benchmarkName":"AIME 2026","benchmarkVersionHint":"aime-2026@v1","metricId":"score","value":94.1,"rawValue":94.1,"unit":"percent","evidenceLevel":"C","sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourceLocator":"rank=13;row=12","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"bd8f3e47321d266bfd4c817736e5f0d18cbcd5500ebef2c3b98cffdf31777d0c","protocol":{"filename":".eval_results/aime_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-27B"},{"id":"pub-66bdd1d311b1b4ea5433c495","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"Qwen/Qwen3.6-35B-A3B","benchmarkId":"aime-2026","benchmarkName":"AIME 2026","benchmarkVersionHint":"aime-2026@v1","metricId":"score","value":92.7,"rawValue":92.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourceLocator":"rank=17;row=16","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"bd8f3e47321d266bfd4c817736e5f0d18cbcd5500ebef2c3b98cffdf31777d0c","protocol":{"filename":".eval_results/aime_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-35B-A3B"},{"id":"pub-e603f3b5727cc0aaceb3eb93","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"moonshotai/Kimi-K2.5","benchmarkId":"aime-2026","benchmarkName":"AIME 2026","benchmarkVersionHint":"aime-2026@v1","metricId":"score","value":95.83,"rawValue":95.83,"unit":"percent","evidenceLevel":"C","sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourceLocator":"rank=6;row=5","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"bd8f3e47321d266bfd4c817736e5f0d18cbcd5500ebef2c3b98cffdf31777d0c","protocol":{"filename":".eval_results/MathArena--aime_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://matharena.ai/?comp=aime--aime_2026"},{"id":"pub-9a29ef28c0d2b42badd8191b","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"moonshotai/Kimi-K2.6","benchmarkId":"aime-2026","benchmarkName":"AIME 2026","benchmarkVersionHint":"aime-2026@v1","metricId":"score","value":96.4,"rawValue":96.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourceLocator":"rank=4;row=3","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"bd8f3e47321d266bfd4c817736e5f0d18cbcd5500ebef2c3b98cffdf31777d0c","protocol":{"filename":".eval_results/aime_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6"},{"id":"pub-0a320e47ff1a3022741ca6c6","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"zai-org/GLM-5","benchmarkId":"aime-2026","benchmarkName":"AIME 2026","benchmarkVersionHint":"aime-2026@v1","metricId":"score","value":95.83,"rawValue":95.83,"unit":"percent","evidenceLevel":"C","sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourceLocator":"rank=7;row=6","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"bd8f3e47321d266bfd4c817736e5f0d18cbcd5500ebef2c3b98cffdf31777d0c","protocol":{"filename":".eval_results/MathArena--aime_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://matharena.ai/?comp=aime--aime_2026"},{"id":"pub-11581d402e03644240b4d374","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"zai-org/GLM-5.1","benchmarkId":"aime-2026","benchmarkName":"AIME 2026","benchmarkVersionHint":"aime-2026@v1","metricId":"score","value":95.3,"rawValue":95.3,"unit":"percent","evidenceLevel":"C","sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","sourceLocator":"rank=11;row=10","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"bd8f3e47321d266bfd4c817736e5f0d18cbcd5500ebef2c3b98cffdf31777d0c","protocol":{"filename":".eval_results/MathArena--aime_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.1"},{"id":"pub-39dc9579246ff50a09053200","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"Claude Fable 5 (High)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.11615315314960728,"rawValue":0.11615315314960728,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=2","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6b6c74c43ad6f325eef8de24","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"Claude Opus 4.6","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.05775102458273259,"rawValue":0.05775102458273259,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=15","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a07e44f7b75ab7a1f5cb7ce3","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"Claude Opus 4.7","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.07179330480418356,"rawValue":0.07179330480418356,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=9","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-0ab61d1006c2cead37f70847","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"Claude Opus 4.7 (High)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.0727341948352638,"rawValue":0.0727341948352638,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=8","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-36248fa8a206907425cf3b40","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"Claude Opus 4.8","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.0271912872310097,"rawValue":0.0271912872310097,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=22","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-185586c3a17d46438ae69692","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"Claude Opus 4.8 (High)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.09988008358694944,"rawValue":0.09988008358694944,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=4","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a23676fdea97d6886ea58468","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"Claude Opus 5 (High)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.12733236681052468,"rawValue":0.12733236681052468,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=0","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-775a2442eff50a6375fd7188","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"Claude Opus 5 (Max)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.12413615325188568,"rawValue":0.12413615325188568,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=1","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-0a600fc00ac1f1e306ee523f","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"Claude Sonnet 4.6","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.022080938177991297,"rawValue":0.022080938177991297,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=23","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-38098b70b75d832bdd94f629","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"Claude Sonnet 5 (High)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.08045201676788406,"rawValue":0.08045201676788406,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=7","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1beefbd1ac64e5f758bef238","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"DeepSeek V4 Pro","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.0018080733374592028,"rawValue":-0.0018080733374592028,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=29","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2095fa947d919162e67a1653","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"DeepSeek V4 Pro (High) (0813)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.06970690759579298,"rawValue":0.06970690759579298,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=10","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-aaa1cee900f2b65fc620748d","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"Deepseek V4 Flash (High) (20260731)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.0346379679398035,"rawValue":0.0346379679398035,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=20","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1eb80a16affa6f47fd3f4c89","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"GLM 5.1","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.004246206481093473,"rawValue":-0.004246206481093473,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=31","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-95dc89b59c846e61d36a6d3c","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"GLM 5.2 (Max)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.05862951282046114,"rawValue":0.05862951282046114,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=14","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d38a265d501d169e799248ef","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"GLM 5.3 (Max)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.03774351573659967,"rawValue":0.03774351573659967,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=19","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-aa680db93269d7e5d51e4500","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"GPT 5.4 (High)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.045811011871245894,"rawValue":0.045811011871245894,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=17","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1d29226e7ab56b9b62bc42bf","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"GPT 5.5","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.056818461920492905,"rawValue":0.056818461920492905,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=16","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-233911bc4dfe5e603d3d94dc","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"GPT 5.5 (High)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.06758524664177534,"rawValue":0.06758524664177534,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=11","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-73325ffdf060eabbec264654","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"GPT 5.5 (xHigh)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.0855784909898381,"rawValue":0.0855784909898381,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=6","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-41ceb54beb7d732578f2af3f","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"GPT 5.6 Luna (xHigh)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.03829833242271487,"rawValue":0.03829833242271487,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=18","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-28961e9a3695953a508666ca","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"GPT 5.6 Sol (xHigh)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.10308907749743668,"rawValue":0.10308907749743668,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=3","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f63e920b5a207f21d088a53e","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"GPT 5.6 Terra (xHigh)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.03298223715667077,"rawValue":0.03298223715667077,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=21","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-37b44338a1cbefddbd1dedd4","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"Gemini 3.1 Pro Preview","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.024091974587494024,"rawValue":-0.024091974587494024,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=35","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-297fe35d70467c0afff797cb","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"Gemini 3.5 Flash (High)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.02221800402823942,"rawValue":-0.02221800402823942,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=34","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-24bedef68563fb85d35e1f9d","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"Gemini 3.5 Flash (Medium)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.03484727666575396,"rawValue":-0.03484727666575396,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=39","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-174cd2be9c579ed18b507ab5","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"Gemini 3.5 Flash Lite","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.10872946128446538,"rawValue":-0.10872946128446538,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=48","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d027fab2c1ae71dd4dc81a1e","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"Gemini 3.6 Flash (High)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.0321966465690072,"rawValue":-0.0321966465690072,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=38","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3f6f3479d5a9d7524ca78092","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"Gemini 3.7 Flash (High)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.015668534602611646,"rawValue":0.015668534602611646,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=26","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2adf258d87b8ea738b3bdba7","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"Grok 4.5","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.06384994584206044,"rawValue":0.06384994584206044,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=13","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fffa0dfb5f17c783f8eebc75","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"Kimi K2.6","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.010858661242663967,"rawValue":0.010858661242663967,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=27","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7961fe5c7ea5159eb206c8ed","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"Kimi K2.7 Code","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.020284108795353084,"rawValue":0.020284108795353084,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=24","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9ef4b938deda4522609c9f3c","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"Kimi K3 (Max)","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.09531968013602259,"rawValue":0.09531968013602259,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=5","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1f43b0d04e214fe8bed28f2b","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"Mimo V2.5 Pro","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.02787033746291722,"rawValue":-0.02787033746291722,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=36","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1a2ee37e073afa42c53b89c6","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"Minimax M2.7","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.1015768677145034,"rawValue":-0.1015768677145034,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=46","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d6685c16766adf319d91d735","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"Minimax M3","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.031007512966943346,"rawValue":-0.031007512966943346,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=37","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-16122c25fd287e99a4e1a9aa","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"Qwen3.7 Max","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.010807488639950926,"rawValue":-0.010807488639950926,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=32","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f8554e026ed13e81144f91f3","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"Qwen3.7 Plus","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":-0.020675286771546914,"rawValue":-0.020675286771546914,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=33","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-95b34f738cacc70ed76fe387","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"Qwen3.8 Max","benchmarkId":"arena-agent","benchmarkName":"Agent Arena","benchmarkVersionHint":"arena-agent@latest","metricId":"ips","value":0.06459728013665357,"rawValue":0.06459728013665357,"unit":"fraction","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=12","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"agent","category":"overall","harness":"arena-agent","rating_method":"ips","split":"latest","subject_type":"system"},"harnessId":"arena-agent","qualityFlags":["agent_ips","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4df5941c9e64215e671e3c62","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1504.2491537844257,"rawValue":1504.2491537844257,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=3","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-69cba88238aba653266e586b","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1423.0272414121418,"rawValue":1423.0272414121418,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=29","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c5aada10249c7222dec885ce","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1461.6531335915963,"rawValue":1461.6531335915963,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=18","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-997a857cf88bb99c7af5e904","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1509.9436313705492,"rawValue":1509.9436313705492,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=1","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-663aca0565d1905fe272e17d","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1505.9121170864287,"rawValue":1505.9121170864287,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=2","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-ac3c6df57a50e69c0a42154d","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1498.1450675830365,"rawValue":1498.1450675830365,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=4","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-aadf5e721d95227f8da1de7c","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-thinking","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1497.0210273683192,"rawValue":1497.0210273683192,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=5","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-fa0293f3d521f4efc7b6f770","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1468.9024272352567,"rawValue":1468.9024272352567,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=15","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-492cf215db25aaa017945e21","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-thinking","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1474.5889982256267,"rawValue":1474.5889982256267,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=11","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c932546fbf518850605ce7e5","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1520.1085960110886,"rawValue":1520.1085960110886,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=0","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-0b53fafca7877cc9bdb559c7","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1446.1948727501483,"rawValue":1446.1948727501483,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=21","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-cd8d406f9107c4f27bfb49cb","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1482.7958418713733,"rawValue":1482.7958418713733,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=7","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-50b582c40453395af34f0acf","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1470.2233955126517,"rawValue":1470.2233955126517,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=13","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-230ef73a7dc629b18a89ba84","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1420.8717297728372,"rawValue":1420.8717297728372,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=30","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-221c089ad7f939a9b220faa2","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1445.3243879762572,"rawValue":1445.3243879762572,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=22","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-7a8d7bec3cb2ad8dd74239b9","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1464.646724197627,"rawValue":1464.646724197627,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=16","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-d036ec92bc673ed59864ab9b","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1417.1818912745823,"rawValue":1417.1818912745823,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=31","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-ce6998704b46d0b27233b594","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1400.8394066066326,"rawValue":1400.8394066066326,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=36","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-94c6a8bf8bb5500e57e7de14","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1400.0724185714537,"rawValue":1400.0724185714537,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=37","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-d9b6d59cf14ec666793c9db7","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1405.0199151746874,"rawValue":1405.0199151746874,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=34","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-62aa69a9861b052029162d41","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1470.1387900819968,"rawValue":1470.1387900819968,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=14","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-83a12c0d67d9272abf955576","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1480.2983593519762,"rawValue":1480.2983593519762,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=8","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-a8d48587c4d6b3f0309265f3","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1484.7040517486435,"rawValue":1484.7040517486435,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=6","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-b4387ff2af571950f29f48a2","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1401.596483346334,"rawValue":1401.596483346334,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=35","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-325f4b41894a283d531b6fc5","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1462.304000455754,"rawValue":1462.304000455754,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=17","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-307c16e3da4c7ade01749734","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1478.6820927474027,"rawValue":1478.6820927474027,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=10","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-66deaf875306c705d625770a","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1479.2063866025728,"rawValue":1479.2063866025728,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=9","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-e264b6734df427514f3bf02a","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1453.732648558952,"rawValue":1453.732648558952,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=19","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-67af861372a31bd3aac819ca","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1429.4883922629656,"rawValue":1429.4883922629656,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=27","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8d21a454c47e939ad50e5e86","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1450.9009906110664,"rawValue":1450.9009906110664,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=20","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-e3cbd881eb2946fd3c33783f","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1434.8163626138064,"rawValue":1434.8163626138064,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=25","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c261d4bfae667a61037f8d65","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-document","benchmarkName":"Arena Document","benchmarkVersionHint":"arena-document@latest","metricId":"arena_score_bt","value":1439.5689281931259,"rawValue":1439.5689281931259,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=24","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-07-30","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"document","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-08dcdd8ac4eb7c6249bdd972","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1230.2225243963933,"rawValue":1230.2225243963933,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=4","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-44215c724c2d3e739e854421","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1179.6445104274662,"rawValue":1179.6445104274662,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=18","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8880b1a192080bc08d70e837","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1253.415899373882,"rawValue":1253.415899373882,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=1","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-01521887ac72987f26e403ef","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1233.2545271621282,"rawValue":1233.2545271621282,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=3","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-745521a5bf4342fbbcfd5fec","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1204.3009106749619,"rawValue":1204.3009106749619,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=11","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-78ef03ff10eddcf9033a34b7","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1126.4656766180015,"rawValue":1126.4656766180015,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=31","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-0c82d64fe143a67d13def394","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1158.4558124801003,"rawValue":1158.4558124801003,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=23","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-152575b3fc26f7a1bcc301f9","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1221.231466167688,"rawValue":1221.231466167688,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=6","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-627a682e57754a3ab9184e1a","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1194.0719510316353,"rawValue":1194.0719510316353,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=16","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-16571658850c79f16d3ab777","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1132.767611287441,"rawValue":1132.767611287441,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=29","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-4303f4a77031da3a1c52ae0a","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1199.4437910533609,"rawValue":1199.4437910533609,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=13","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-5879cc04bd3ca6b06e7e03e2","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1206.9460713540655,"rawValue":1206.9460713540655,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=10","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-1c1b4ed72231ea0c7e8d2a07","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-search-non-reasoning","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1172.1381005441356,"rawValue":1172.1381005441356,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=19","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-6a996e7720c320a1fd8c0f64","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1197.1574686457088,"rawValue":1197.1574686457088,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=15","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-464bf310c1b82f8210623022","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1242.052705813796,"rawValue":1242.052705813796,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=2","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8070d426b107e80ee37239be","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1257.2603500980206,"rawValue":1257.2603500980206,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=0","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-4ec565bbd96b3bf02151d41a","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1170.50892593241,"rawValue":1170.50892593241,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=21","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8a08ce7f85de876f893e75cb","canonicalModelId":"xai/grok-4@current","modelName":"Grok 4","modelRef":"grok-4-search","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1141.5180851711918,"rawValue":1141.5180851711918,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=27","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-44330158b5ab52e827070332","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-search","benchmarkName":"Arena Search","benchmarkVersionHint":"arena-search@latest","metricId":"arena_score_bt","value":1212.646185108397,"rawValue":1212.646185108397,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=7","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-24","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"search","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-565200f47136ca2e363ffedc","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1494.684788253507,"rawValue":1494.684788253507,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=4","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-4d73795515b66664dabd3106","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1519.2356478977404,"rawValue":1519.2356478977404,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=762","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-a49a6476a6ba1ea442b13cd4","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1544.239298121939,"rawValue":1544.239298121939,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=400","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-26623cbd8e3cf1439abec5c7","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1452.770283211219,"rawValue":1452.770283211219,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=839","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-46083043d86a2d2e9a68e838","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1394.8817158702782,"rawValue":1394.8817158702782,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=142","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-f19ff3eda0eb3f3e4d9c4098","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1422.9434261424383,"rawValue":1422.9434261424383,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=537","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-ae688607720557f54688f08a","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1401.6403101454246,"rawValue":1401.6403101454246,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=910","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-07a03b229d31ff4b4c531591","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1385.831066547008,"rawValue":1385.831066547008,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=560","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-ed65d4db95824ba5f513b84f","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1364.290818279198,"rawValue":1364.290818279198,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=173","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-d1a1e1c53f71d4a3ebaf3eb9","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514-thinking-16k","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1442.6087338228988,"rawValue":1442.6087338228988,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=853","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-08dcc60a1364a553162d991c","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514-thinking-16k","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1375.7212726765192,"rawValue":1375.7212726765192,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=157","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-c12e0775113004d87dfb27b2","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514-thinking-16k","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1385.69489723648,"rawValue":1385.69489723648,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=561","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-934c49bb50d96878a2767dc4","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1498.4999714495004,"rawValue":1498.4999714495004,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=772","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-3fb626b844ded292d9debc57","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1449.8532784846016,"rawValue":1449.8532784846016,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=41","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-159c38514aa5a59834b2d028","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1473.9154377418088,"rawValue":1473.9154377418088,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=463","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-43131b87e65c86e9f1e5b1b9","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-high-32k","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1502.334287039247,"rawValue":1502.334287039247,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=769","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-e7f4985d9a7e979eca7405ea","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-high-32k","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1446.6977054302,"rawValue":1446.6977054302,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=44","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-3dcbafb79e1decd664359400","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-high-32k","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1461.0518393877824,"rawValue":1461.0518393877824,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=486","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-72ddbb43542525d82b165399","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1533.8798441359052,"rawValue":1533.8798441359052,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=759","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-3c598c07cd2694769b42625f","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1497.1950053437704,"rawValue":1497.1950053437704,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=3","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-362bc5b596fdc66ba3cfbb7c","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1550.6343188146861,"rawValue":1550.6343188146861,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=398","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-efd0eed446610099febd250d","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1535.4160363275219,"rawValue":1535.4160363275219,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=758","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-5d68a4f3bbc38896ed4f9d4d","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1502.6783117818993,"rawValue":1502.6783117818993,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=2","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-fc3e03c69a1de251f0524ade","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1554.5618817134211,"rawValue":1554.5618817134211,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=397","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-2a3f29f4c525c0fb53e39877","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1515.6750721273409,"rawValue":1515.6750721273409,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=764","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-42c37695164fbd21f8b13c87","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1482.7423547986384,"rawValue":1482.7423547986384,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=8","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-25b5e287a2ed3c3e5707ebae","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1528.1108979767837,"rawValue":1528.1108979767837,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=412","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-1eeb61d4e8c05e901da04dc8","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1519.1148988655398,"rawValue":1519.1148988655398,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=763","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-ca0d0d4b1bed486d3d53895b","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1490.1204074818936,"rawValue":1490.1204074818936,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=6","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-00f82ba42564ac839c99b284","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1534.9680371881902,"rawValue":1534.9680371881902,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=405","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-b05e539994a1c0ba73efccdf","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1484.3170781766582,"rawValue":1484.3170781766582,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=791","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-915cda4455cdadd9cf5eb6e0","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1505.2729249443958,"rawValue":1505.2729249443958,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=429","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-8077b814799e387eddaf3ce6","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1451.946373879298,"rawValue":1451.946373879298,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=37","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-82cc33023b255da2fca0525d","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1489.6262287811883,"rawValue":1489.6262287811883,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=781","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-af66a9a956dd795efd57e2ae","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1460.9095664685228,"rawValue":1460.9095664685228,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=29","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-86b47fb3c0c7a3b5f13f36ca","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1501.446206226374,"rawValue":1501.446206226374,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=434","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-ce323b75362323b431e08cf4","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1504.2025661543232,"rawValue":1504.2025661543232,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=0","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-7194ca77d2c320b080ee4b92","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1585.0807825560232,"rawValue":1585.0807825560232,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=395","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-fc71e936af09b840ae5d92e5","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1531.5185073420098,"rawValue":1531.5185073420098,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=760","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-f058636aceda8a0937217553","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1588.2900384300049,"rawValue":1588.2900384300049,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=394","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-025b1677857c3f8d24e5f7b9","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1504.0374724478093,"rawValue":1504.0374724478093,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=1","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-6f180f0c11efaff98f0ddaf6","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1528.395960285115,"rawValue":1528.395960285115,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=761","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-641888df0bdb73508d75ce61","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1485.4495918725258,"rawValue":1485.4495918725258,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=789","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-93142edb4299a5096ca4d289","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1437.9788159129344,"rawValue":1437.9788159129344,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=67","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-c5a05e48e56e4984bf94ed02","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1466.401517747521,"rawValue":1466.401517747521,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=479","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-f8b61cac5d32b76b1bce5246","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929-high-32k","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1488.2099563202128,"rawValue":1488.2099563202128,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=783","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c8fae60c2be1e049937019a0","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929-high-32k","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1433.3331108484715,"rawValue":1433.3331108484715,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=77","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-ddfc58e95078025ad9dc5f32","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929-high-32k","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1467.1058479923634,"rawValue":1467.1058479923634,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=476","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-4aa491b283cc2441a56eeed6","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1503.7881653396375,"rawValue":1503.7881653396375,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=767","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-9408ab2bc22ddde1d591c100","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1457.5719381664967,"rawValue":1457.5719381664967,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=30","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-f420cdd350a8fa9f9f147244","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1492.7292671098887,"rawValue":1492.7292671098887,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=444","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-01ed006de364bf1b453f1120","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1485.463794561637,"rawValue":1485.463794561637,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=788","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-76f6286e17d2c4a25002509b","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1442.2497390160604,"rawValue":1442.2497390160604,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=56","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-2b2b03f55fcbaf0a6bb08407","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1482.3095417342138,"rawValue":1482.3095417342138,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=453","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-9f5d591e0c16975be8032b60","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"deepseek-r1-0528","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1427.9090691139086,"rawValue":1427.9090691139086,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=84","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-877652b38949a49d7885f0ad","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"deepseek-r1-0528","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1456.7251609405107,"rawValue":1456.7251609405107,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=495","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-d7a32b09bf13a6e713c88e72","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"deepseek-r1-0528","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1427.0808103160361,"rawValue":1427.0808103160361,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=879","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-55434b64fe5970ef4f6b5f2c","canonicalModelId":"deepseek/v3.1@2025-08-21","modelName":"DeepSeek V3.1","modelRef":"deepseek-v3.1","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1455.8103764607301,"rawValue":1455.8103764607301,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=497","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-55ac59e3cc366f57f11b5eab","canonicalModelId":"deepseek/v3.1@2025-08-21","modelName":"DeepSeek V3.1","modelRef":"deepseek-v3.1","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1419.1240636772848,"rawValue":1419.1240636772848,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=104","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-68a67546179f1603804b79d7","canonicalModelId":"deepseek/v3.1@2025-08-21","modelName":"DeepSeek V3.1","modelRef":"deepseek-v3.1","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1414.748302689449,"rawValue":1414.748302689449,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=892","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-b9c2b7775022246baa75f161","canonicalModelId":"deepseek/v3.1-terminus@2025-09-22","modelName":"DeepSeek V3.1 Terminus","modelRef":"deepseek-v3.1-terminus","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1416.6386560476492,"rawValue":1416.6386560476492,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=111","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-9cef733c8963501d396fb16f","canonicalModelId":"deepseek/v3.1-terminus@2025-09-22","modelName":"DeepSeek V3.1 Terminus","modelRef":"deepseek-v3.1-terminus","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1406.5948738610255,"rawValue":1406.5948738610255,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=905","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-6985f355e39c9663c64c3d3d","canonicalModelId":"deepseek/v3.1-terminus@2025-09-22","modelName":"DeepSeek V3.1 Terminus","modelRef":"deepseek-v3.1-terminus-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1419.5842644243733,"rawValue":1419.5842644243733,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=101","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8e1bc949df10eca0f45d2453","canonicalModelId":"deepseek/v3.1-terminus@2025-09-22","modelName":"DeepSeek V3.1 Terminus","modelRef":"deepseek-v3.1-terminus-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1428.5142413891792,"rawValue":1428.5142413891792,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=876","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-3ad657921e7ebbf4d3afee21","canonicalModelId":"deepseek/v3.1@2025-08-21","modelName":"DeepSeek V3.1","modelRef":"deepseek-v3.1-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1470.4827142810077,"rawValue":1470.4827142810077,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=468","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c68e9a12999e78898cd7b7f6","canonicalModelId":"deepseek/v3.1@2025-08-21","modelName":"DeepSeek V3.1","modelRef":"deepseek-v3.1-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1416.5083613666134,"rawValue":1416.5083613666134,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=112","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-68145544e60fa37a8f76eeec","canonicalModelId":"deepseek/v3.1@2025-08-21","modelName":"DeepSeek V3.1","modelRef":"deepseek-v3.1-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1418.4119018236506,"rawValue":1418.4119018236506,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=888","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-6fd7aeac36e0e211ac94608a","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1449.0826070625546,"rawValue":1449.0826070625546,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=843","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-411f073f4bad4eaef70314d5","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1424.6209587565081,"rawValue":1424.6209587565081,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=89","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-f37589e083103561ae884011","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1459.500925724597,"rawValue":1459.500925724597,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=489","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-690c6e7af75562e94f9ddb70","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"deepseek-v3.2-exp","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1464.8964135646193,"rawValue":1464.8964135646193,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=482","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8f11bae78f0c00350144c336","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"deepseek-v3.2-exp","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1422.5513334398902,"rawValue":1422.5513334398902,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=92","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-62e5ea5a0decc3f8fbae1e61","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"deepseek-v3.2-exp","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1433.403071405,"rawValue":1433.403071405,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=870","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-bf58c6356f5f047a20a04cf9","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"deepseek-v3.2-exp-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1424.3777108565635,"rawValue":1424.3777108565635,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=90","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-fd771c59e87c584d83fa4076","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"deepseek-v3.2-exp-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1459.9497966396993,"rawValue":1459.9497966396993,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=488","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-47585a8666b61e994259f99f","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"deepseek-v3.2-exp-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1438.6471525701834,"rawValue":1438.6471525701834,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=861","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-fb5332cac6fe50d0ccd2674b","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1453.271826663273,"rawValue":1453.271826663273,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=837","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c003bd58ccffad0baef4b233","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1419.8045027712699,"rawValue":1419.8045027712699,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=100","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-ec079d7853253e47b14f0a8e","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1456.661947443516,"rawValue":1456.661947443516,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=496","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-a2d2885daa92fd7244336d1a","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1455.186846872624,"rawValue":1455.186846872624,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=832","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-3e0a5d5ef4cebf7c700f93e6","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1431.6067430969854,"rawValue":1431.6067430969854,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=78","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-40f90948a1b72b9df334324e","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1466.6123503314598,"rawValue":1466.6123503314598,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=478","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-45d4161fafd19e6e4bdd4adb","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-high-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1468.62205505987,"rawValue":1468.62205505987,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=472","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-d7323e6ab389102847a54fcc","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-high-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1423.9732661575526,"rawValue":1423.9732661575526,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=91","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-dd4b8e012760d4cc28e0c3b4","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-high-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1439.749595500906,"rawValue":1439.749595500906,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=858","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-f988ba94f5056000664c5e7f","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1450.8668626687377,"rawValue":1450.8668626687377,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=39","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-b8b5227669dd4a5f2f8b61cd","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1470.4607075830108,"rawValue":1470.4607075830108,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=812","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-5c8702185cacc3656f9645f2","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1486.8871828117715,"rawValue":1486.8871828117715,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=448","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-1c2e719e8b7c2e625406c7b3","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-high-20260813","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1472.1368816979355,"rawValue":1472.1368816979355,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=810","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-61f2e3f4655654084d677ba4","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-high-20260813","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1484.4183615633667,"rawValue":1484.4183615633667,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=451","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-b08151d26da31dd6ced12cdf","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-high-20260813","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1439.1643609270914,"rawValue":1439.1643609270914,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=63","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-2e21cfa386f45e3d0cd4ee93","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro-high-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1444.9973961807061,"rawValue":1444.9973961807061,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=49","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-3470a0222e0fcfd070765cd7","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro-high-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1484.7833319190365,"rawValue":1484.7833319190365,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=450","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-efefe0b8e0841cd2c95e35cb","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro-high-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1453.6216111078786,"rawValue":1453.6216111078786,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=834","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-c38c7e0d690399494980a06a","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1417.2991182812218,"rawValue":1417.2991182812218,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=110","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-80279be55a468b3d555a14be","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1450.9718021708911,"rawValue":1450.9718021708911,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=505","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-abf046af69e4600ffb328663","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1423.5268495127216,"rawValue":1423.5268495127216,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=884","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-7b6382051ac174ed36c0a5e6","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1457.3391897717306,"rawValue":1457.3391897717306,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=31","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-47eef41792434a5dec789e29","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1506.8141348240842,"rawValue":1506.8141348240842,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=428","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-f35df1c9b13b313be505ed46","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1452.4271145093594,"rawValue":1452.4271145093594,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=840","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-f392a193cb25322717236799","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1479.5500212955633,"rawValue":1479.5500212955633,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=11","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-59a93cd31c9d8da683ee4781","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1531.4621627000524,"rawValue":1531.4621627000524,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=408","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-54a58737f240c9d62b434c22","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1483.3184123763037,"rawValue":1483.3184123763037,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=794","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-7dc86d27ec9ed2ab603f2f21","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1537.6543467895315,"rawValue":1537.6543467895315,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=402","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-b394a3c1fb05e562c2c3c894","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1482.6325665947309,"rawValue":1482.6325665947309,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=9","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-ecf366af57b815b7c4f7723e","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1491.9294583881592,"rawValue":1491.9294583881592,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=779","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-5337677dc70f7f99d0020522","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1474.565145064679,"rawValue":1474.565145064679,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=462","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-e066d0d3925d61eb84f1357a","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1436.4808617125807,"rawValue":1436.4808617125807,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=72","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-251b2ab681acc008f82b0372","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1451.4549706628025,"rawValue":1451.4549706628025,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=842","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-bf4bc906a94c11d683ee76b4","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1475.206155988065,"rawValue":1475.206155988065,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=17","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-a11cb0134d5ac156eb69e07d","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1520.2097875893496,"rawValue":1520.2097875893496,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=416","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-01a1b0ba30809c8d881f3ad3","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1487.6627859976325,"rawValue":1487.6627859976325,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=785","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-a7830f3d08c7a8a1a278346a","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1541.8997963210757,"rawValue":1541.8997963210757,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=401","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-e188a93a68e303167347a67c","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1476.4838647133247,"rawValue":1476.4838647133247,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=15","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-c943823991df56be9fb045a4","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1491.6956027567426,"rawValue":1491.6956027567426,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=780","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-44eeb791528061b6d175796c","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1562.1742490953136,"rawValue":1562.1742490953136,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=396","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-2ae748d37d67edbee0966038","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1490.2116753124994,"rawValue":1490.2116753124994,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=5","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-cdc5b0bdf5c15ce1d8ca5e9c","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1503.5280062667107,"rawValue":1503.5280062667107,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=768","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-7a8c7ee54ad9b2dea12575da","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1485.77033925055,"rawValue":1485.77033925055,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=449","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-5fed400a773ef90f7c49b98f","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1455.7302170826042,"rawValue":1455.7302170826042,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=831","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-dcb858251fbad1bde1658385","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1435.3354115275959,"rawValue":1435.3354115275959,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=74","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-9b18543e3f310f00b4ce05df","canonicalModelId":"zai/glm-4.7-flash@current","modelName":"GLM-4.7 Flash","modelRef":"glm-4.7-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1424.4551258868103,"rawValue":1424.4551258868103,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=534","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-0948ab4e0c8f561432646fd8","canonicalModelId":"zai/glm-4.7-flash@current","modelName":"GLM-4.7 Flash","modelRef":"glm-4.7-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1383.852604453216,"rawValue":1383.852604453216,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=926","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-4063483a1830c1dfbbae32df","canonicalModelId":"zai/glm-4.7-flash@current","modelName":"GLM-4.7 Flash","modelRef":"glm-4.7-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1352.9259898135635,"rawValue":1352.9259898135635,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=183","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-3bd3bd9a99264121e065e395","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1514.545958762338,"rawValue":1514.545958762338,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=423","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-bd23eab86d49b31dacb8cbb8","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1445.2412180980982,"rawValue":1445.2412180980982,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=47","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-d4a7be1bca5a8f2c6d03b256","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1460.3721655134777,"rawValue":1460.3721655134777,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=826","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-74dd6eb5d8ad21f557ccc9f8","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1520.10410144335,"rawValue":1520.10410144335,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=417","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-ddf00a21f6e433fda4d5e6bb","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1488.1012064231313,"rawValue":1488.1012064231313,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=784","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-e51613ab4f5a9122e0a00b8f","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1464.107000003788,"rawValue":1464.107000003788,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=28","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-304bb6e78630965076e8053a","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1516.3461367724533,"rawValue":1516.3461367724533,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=420","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-1bb3b53f8d78dfec7f0765e8","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1465.3786574650403,"rawValue":1465.3786574650403,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=26","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-dbc642135df06b48e9aad3bf","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1475.473869793419,"rawValue":1475.473869793419,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=803","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-8b0af01ee73e845754f6dc84","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1504.0079599546075,"rawValue":1504.0079599546075,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=766","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-f9467d9b5c03450f4b272eb4","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1531.996115754834,"rawValue":1531.996115754834,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=407","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-eb5791f1973a201cd2a1169c","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1476.929093287472,"rawValue":1476.929093287472,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=14","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-c1ff19d552257ee3521cf9a8","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1492.9304481628153,"rawValue":1492.9304481628153,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=443","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-103558295d27ac7eb267f86d","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1468.005293829955,"rawValue":1468.005293829955,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=815","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-775ccc0bc001838bf3126be6","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1436.7987275161317,"rawValue":1436.7987275161317,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=70","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-2d498bd59b01a6033c89d14f","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1435.5240208841885,"rawValue":1435.5240208841885,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=866","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-69da1b7ae2bb86bc4f41450d","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1405.633007885001,"rawValue":1405.633007885001,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=130","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-1c9d362f062c5db37ac39993","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1423.687412475318,"rawValue":1423.687412475318,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=536","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-a3570b0debd3c747515cec65","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1406.7245246472921,"rawValue":1406.7245246472921,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=904","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-2d027360a2881ed477d77911","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1373.4392475339996,"rawValue":1373.4392475339996,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=160","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-f82cde7143b716ddb1ea69df","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1386.3329871664873,"rawValue":1386.3329871664873,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=559","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-279641ec268c28b2569053fc","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1349.6979522075699,"rawValue":1349.6979522075699,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=584","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-2c45441a73d732fb9e72a31c","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1352.4736004900342,"rawValue":1352.4736004900342,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=960","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-0a90630ce620b718ee0f4c2b","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1320.2673210882358,"rawValue":1320.2673210882358,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=211","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-dba3ae933c6fe4f76855c264","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1474.9413099648884,"rawValue":1474.9413099648884,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=460","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-d18043e5964127d79b38bd76","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1422.2190385742006,"rawValue":1422.2190385742006,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=94","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-b0cf7ee9862679ed54a188f7","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1436.3313060625342,"rawValue":1436.3313060625342,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=864","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-f83b7e31be61501fd2ddbd29","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1495.4011508648664,"rawValue":1495.4011508648664,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=440","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-b0a3e236ae308f62b85665e8","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1441.3712036369134,"rawValue":1441.3712036369134,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=58","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-7c36f015c51f948b52741790","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1452.413214579112,"rawValue":1452.413214579112,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=841","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-2822f01ac746e1e15cbba927","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1431.7860968982502,"rawValue":1431.7860968982502,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=873","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-ca9fe72a445ff817d5559cd8","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1449.207461053533,"rawValue":1449.207461053533,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=511","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-c345d04a0f1891bec2ad9c2f","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1411.9324804584382,"rawValue":1411.9324804584382,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=120","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-6349aea34a82d615e06c77ed","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1442.4392042103425,"rawValue":1442.4392042103425,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=854","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-e306fda394d1f9ebec961faf","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1450.6337964584836,"rawValue":1450.6337964584836,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=507","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-12d4c284738adaba05fcecc4","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1416.4915639243197,"rawValue":1416.4915639243197,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=113","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-f0d6554c940c1738aff26dd9","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1509.7354893174718,"rawValue":1509.7354893174718,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=426","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-3b376d053d6a02967cda0f0d","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1452.5898889549094,"rawValue":1452.5898889549094,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=35","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-5f1e69a918f0d82b7c22deb4","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1480.1373965230498,"rawValue":1480.1373965230498,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=799","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-4b9b551f0147b600d21ba83d","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1495.808303356952,"rawValue":1495.808303356952,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=774","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-527017855db531446e9c3460","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1469.5687129457133,"rawValue":1469.5687129457133,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=22","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-7285aefab3599a081162bdba","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1518.363983589094,"rawValue":1518.363983589094,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=418","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-3e3d85510d16d10211955eb1","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1439.085991867134,"rawValue":1439.085991867134,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=859","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-07f7de66eb9329380124ba86","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1450.0267388282168,"rawValue":1450.0267388282168,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=509","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-43f867c038f73c827e6c08bb","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1412.1044098861782,"rawValue":1412.1044098861782,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=119","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-58873118bc0643218cdf6ee7","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1404.044665924795,"rawValue":1404.044665924795,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=908","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-f8e079db6ac622c9cb5f3d46","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1390.7940153666893,"rawValue":1390.7940153666893,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=555","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-c5c44201f69e931af213227a","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1372.7685819566527,"rawValue":1372.7685819566527,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=161","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-7365d0cc744afee5b9c45231","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1537.421654292945,"rawValue":1537.421654292945,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=403","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-a96192f3263ae998d06b0704","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1465.8000166386585,"rawValue":1465.8000166386585,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=25","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-8020cb7229b3332aa722bf29","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1481.642180792169,"rawValue":1481.642180792169,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=798","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-e9d2dd548bc8e9b49d0965b6","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1494.471152235676,"rawValue":1494.471152235676,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=775","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-8183d6a83c0323728b11fb5b","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1525.3095553781413,"rawValue":1525.3095553781413,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=415","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-3b88a37c10fe03d86c5944ae","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1470.8780077273245,"rawValue":1470.8780077273245,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=21","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-784d0ed138be1e70dc43b425","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1458.0333757430712,"rawValue":1458.0333757430712,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=492","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-15aad823bca8b5f0dcc46d47","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1419.5386838826466,"rawValue":1419.5386838826466,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=102","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-d9b23fffb3bc508cc9fa8e7f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1434.207722401864,"rawValue":1434.207722401864,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=868","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-6c43980bd1293502685ac786","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1463.4279627739224,"rawValue":1463.4279627739224,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=822","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-b4b9c2d0558aef514cc96e5c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1467.224316956766,"rawValue":1467.224316956766,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=475","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-4e84c51d02f6d6552342f9da","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1428.5368364471099,"rawValue":1428.5368364471099,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=82","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-fdfc57aea97e1ef42bb60e19","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1532.2918933515473,"rawValue":1532.2918933515473,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=406","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-0d387c6a952a402ef0a2fc24","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1492.2691050475642,"rawValue":1492.2691050475642,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=778","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-e80ac457acfa32fa40ae33e7","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1454.1709198249014,"rawValue":1454.1709198249014,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=34","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-223dbbe1cc840297b5d2e31d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1515.565205789121,"rawValue":1515.565205789121,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=421","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-cdd3adda03d2748bdb534616","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1484.9891712747985,"rawValue":1484.9891712747985,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=790","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-1e0a12777bf5d864d3da0d57","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1446.0343010991212,"rawValue":1446.0343010991212,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=46","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-8aaef5fc5d193c39594de755","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast-reasoning","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1435.5106103793898,"rawValue":1435.5106103793898,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=520","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-05bb671297e0db857d6b082c","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast-reasoning","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1418.1419413595122,"rawValue":1418.1419413595122,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=889","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-006fc75e82b164b23126af1c","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast-reasoning","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1397.9462902761322,"rawValue":1397.9462902761322,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=138","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-f20ba3dfce7e751fb02a04f6","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1452.2990764053388,"rawValue":1452.2990764053388,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=36","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-79c527640a9c165ce774b2a8","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1483.0229340466567,"rawValue":1483.0229340466567,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=795","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-fbf2d320cd7d1eff898035bf","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1499.7382999269346,"rawValue":1499.7382999269346,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=436","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-efd85b6d9f78f2cbb8383eca","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1529.4649127956318,"rawValue":1529.4649127956318,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=410","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-d62f0ff0dba278ffd9464148","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1474.963430362601,"rawValue":1474.963430362601,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=805","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-8001312bb8c81d7750ffce9c","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6-high","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1443.7345059556712,"rawValue":1443.7345059556712,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=53","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-fb683844046426bb5815dc63","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1484.2317973593354,"rawValue":1484.2317973593354,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=792","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-accf314ad770bd8b30bfc029","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1460.7031659670874,"rawValue":1460.7031659670874,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=487","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-2c19322d53b5d62aa9580f32","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1420.3454023693812,"rawValue":1420.3454023693812,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=98","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-4a0d11c8f5a766da1aabe8a6","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1500.265497589836,"rawValue":1500.265497589836,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=435","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-334ec56652cca04554d1dcee","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1445.2327043266241,"rawValue":1445.2327043266241,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=48","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-9770bd46eeb752dd545daf5a","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1473.2918632879823,"rawValue":1473.2918632879823,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=808","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-ae621c9909ee3492a97363ec","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1529.8696294273634,"rawValue":1529.8696294273634,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=409","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-16d1fe61507ad66a39500bcf","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1486.734867015527,"rawValue":1486.734867015527,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=787","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-7b133655be3c95e292075139","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1455.0168217501182,"rawValue":1455.0168217501182,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=33","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-6e249e7512cce04703715fa6","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1511.541768583301,"rawValue":1511.541768583301,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=765","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-d6953c5fe748882bd871729c","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1476.2899572288309,"rawValue":1476.2899572288309,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=16","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-265fa8abcd885feaee917c54","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1515.482090092387,"rawValue":1515.482090092387,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=422","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-f95b1a65662b1ee212ee93d9","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (non-thinking)","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1466.0034476538067,"rawValue":1466.0034476538067,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=480","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-4338d83a2ee07ff441b736b8","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (non-thinking)","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1440.9802098647078,"rawValue":1440.9802098647078,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=856","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-bbbbf877f6af079c3c67fa23","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (non-thinking)","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1411.1437952958465,"rawValue":1411.1437952958465,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=122","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-27807917fec3611702b025ec","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (thinking)","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1432.6867312156085,"rawValue":1432.6867312156085,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=522","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-b1fbd6d98b97d1cd014e7454","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (thinking)","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1417.2421356877353,"rawValue":1417.2421356877353,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=891","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-9ae1a6a71da6bac1fd313e78","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (thinking)","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1394.6498126524996,"rawValue":1394.6498126524996,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=143","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-149ab517a4ede50b669b838d","canonicalModelId":"xiaomi/mimo-v2-omni@current","modelName":"MiMo V2 Omni","modelRef":"mimo-v2-omni","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1464.5854192460758,"rawValue":1464.5854192460758,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=821","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-36271f70b23971bc94ff243b","canonicalModelId":"xiaomi/mimo-v2-omni@current","modelName":"MiMo V2 Omni","modelRef":"mimo-v2-omni","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1467.756636526635,"rawValue":1467.756636526635,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=474","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-f91b094976c08fd0671f8894","canonicalModelId":"xiaomi/mimo-v2-omni@current","modelName":"MiMo V2 Omni","modelRef":"mimo-v2-omni","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1422.2637647204183,"rawValue":1422.2637647204183,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=93","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-78c9aed4a1c111d3e8c6fc19","canonicalModelId":"xiaomi/mimo-v2-pro@current","modelName":"MiMo V2 Pro","modelRef":"mimo-v2-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1476.424896591302,"rawValue":1476.424896591302,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=802","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-d7beedab01aca0bb16e3a0fd","canonicalModelId":"xiaomi/mimo-v2-pro@current","modelName":"MiMo V2 Pro","modelRef":"mimo-v2-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1436.526656194733,"rawValue":1436.526656194733,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=71","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-be5cfdd2174e7352ea6e8f83","canonicalModelId":"xiaomi/mimo-v2-pro@current","modelName":"MiMo V2 Pro","modelRef":"mimo-v2-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1455.5304295602677,"rawValue":1455.5304295602677,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=498","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-d0e99b2c36455efe5857c514","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1468.9854695540266,"rawValue":1468.9854695540266,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=813","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-3cf9a59e905a3d7455164c6e","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1471.3315083707973,"rawValue":1471.3315083707973,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=466","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-afa14db6221d376b9a23170d","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1427.3345994944987,"rawValue":1427.3345994944987,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=85","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-24989972b2c048ddb00da5f6","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1499.3478828199889,"rawValue":1499.3478828199889,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=770","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-0421a9f2a337516c6f26f816","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1464.96741189912,"rawValue":1464.96741189912,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=27","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-8bc84b9c4388ebbf7e20eb30","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1504.1336126761466,"rawValue":1504.1336126761466,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=431","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-685283696547c9d6612288b8","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"minimax-m2.1-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1423.0252338599134,"rawValue":1423.0252338599134,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=885","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c019694e961d2abb3a01a786","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"minimax-m2.1-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1430.9727612173845,"rawValue":1430.9727612173845,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=525","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-e506d1975fbf75f9b6d3abce","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"minimax-m2.1-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1391.0609391665223,"rawValue":1391.0609391665223,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=147","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-a210c98d0a34a6491809618f","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"minimax-m2.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1388.6823361591046,"rawValue":1388.6823361591046,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=556","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-b65c2a0f09f5930da48f04aa","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"minimax-m2.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1378.5858443722477,"rawValue":1378.5858443722477,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=933","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-32e37231e713d0cf59731729","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"minimax-m2.5","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1358.9609122667448,"rawValue":1358.9609122667448,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=177","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-893fc63bd9f9abe0274510af","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2.7","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1453.4669538933076,"rawValue":1453.4669538933076,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=835","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c99e6370ba6fa74ac23a774e","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2.7","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1437.4985997127415,"rawValue":1437.4985997127415,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=518","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-6928a65c850186392c18b82f","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2.7","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1405.2729700708335,"rawValue":1405.2729700708335,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=131","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-ff5b4bd466261f0dcfea9d1b","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1473.4411759865222,"rawValue":1473.4411759865222,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=806","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-de7fae8ae533cfef859dc3ee","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1434.7629318817199,"rawValue":1434.7629318817199,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=75","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-c651a233f07e3f9586420a18","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1458.0451634068265,"rawValue":1458.0451634068265,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=491","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-7ed15d96a7ac1213e3f73d61","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"qwen3-coder-480b-a35b-instruct","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1412.5711018740342,"rawValue":1412.5711018740342,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=899","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-95e4f11abe79909efc5f2cbf","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"qwen3-coder-480b-a35b-instruct","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1356.5686108842956,"rawValue":1356.5686108842956,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=180","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-2abf0829b8f18f0d99bd3cf4","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"qwen3-coder-480b-a35b-instruct","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1355.1825086820477,"rawValue":1355.1825086820477,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=580","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-6ea801bc49e01e96ed78f9fb","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1463.164696604873,"rawValue":1463.164696604873,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=484","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-76420184ec102a973ea9ecb0","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1436.2856315236722,"rawValue":1436.2856315236722,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=865","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-6067f2b718e5161a4354b4a0","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1417.8843322849618,"rawValue":1417.8843322849618,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=107","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-e000cd7f9c8595d2b2ea44a4","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1475.397041496677,"rawValue":1475.397041496677,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=459","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-e5eab78ada5f861875120ee9","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1424.5979855989967,"rawValue":1424.5979855989967,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=882","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-d001f427c427250d4fee3510","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1407.8616630235158,"rawValue":1407.8616630235158,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=127","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-85bc6854e9d0d779970fb6d5","canonicalModelId":"qwen/qwen3.5-35b-a3b@current","modelName":"Qwen3.5 35B-A3B","modelRef":"qwen3.5-35b-a3b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1464.5961634775433,"rawValue":1464.5961634775433,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=483","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-77569fbb7fa8c07d851250b0","canonicalModelId":"qwen/qwen3.5-35b-a3b@current","modelName":"Qwen3.5 35B-A3B","modelRef":"qwen3.5-35b-a3b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1395.586964212609,"rawValue":1395.586964212609,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=141","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-f9bb31ab9622a6e3729286a0","canonicalModelId":"qwen/qwen3.5-35b-a3b@current","modelName":"Qwen3.5 35B-A3B","modelRef":"qwen3.5-35b-a3b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1409.761729852643,"rawValue":1409.761729852643,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=901","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-8a7041aa9a6e4f8a87dc58a4","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1498.948733077815,"rawValue":1498.948733077815,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=438","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-21cd79b4c6d36d087531b86f","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1464.7670373393412,"rawValue":1464.7670373393412,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=820","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-1acd3f339c4b4a09876a2000","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1438.2532016582525,"rawValue":1438.2532016582525,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=66","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-f6ce8cb49b4f758035419399","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1451.0570878730985,"rawValue":1451.0570878730985,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=504","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8957ec3e6a40e8ef6e87d9b0","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1397.599796544761,"rawValue":1397.599796544761,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=139","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-d7a596923be54022ee8744fc","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1412.8720926205697,"rawValue":1412.8720926205697,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=898","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-feb71eb534cf165a4e5be1da","canonicalModelId":"qwen/qwen3.5-max-preview@2026-03-18","modelName":"Qwen3.5 Max Preview","modelRef":"qwen3.5-max-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1535.6774423499453,"rawValue":1535.6774423499453,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=404","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-3449886053bb184c57112a88","canonicalModelId":"qwen/qwen3.5-max-preview@2026-03-18","modelName":"Qwen3.5 Max Preview","modelRef":"qwen3.5-max-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1470.9246252658847,"rawValue":1470.9246252658847,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=20","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-86f8b9e5fe9631be2305c4ee","canonicalModelId":"qwen/qwen3.5-max-preview@2026-03-18","modelName":"Qwen3.5 Max Preview","modelRef":"qwen3.5-max-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1486.9869932732572,"rawValue":1486.9869932732572,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=786","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-06c8275cb7f0e88f385125a7","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1446.2675776614128,"rawValue":1446.2675776614128,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=45","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-6701af42ce2324130c84fe56","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1490.283888090373,"rawValue":1490.283888090373,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=445","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-98c06a46043d15221a6dab5a","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1467.9806594898603,"rawValue":1467.9806594898603,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=816","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-08f7b68788879ff60a204e68","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1465.9333014962801,"rawValue":1465.9333014962801,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=819","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-171f7eefb88e0f5f028425cb","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1474.8119844446103,"rawValue":1474.8119844446103,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=461","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-c7ba10f12b9a26b992deec6f","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1436.8284922613113,"rawValue":1436.8284922613113,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=69","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-091145f24182ed7e5a9e46bd","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1497.5206352718767,"rawValue":1497.5206352718767,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=773","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-f3bdc729fcb255546c357c29","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1474.2960161967428,"rawValue":1474.2960161967428,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=18","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-deb42559be97f11c66c12929","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max-preview","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1525.6221592269271,"rawValue":1525.6221592269271,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=414","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-03f8b2e206df5847da0f8840","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1456.2401150516928,"rawValue":1456.2401150516928,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=32","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-cf3b8e98f05a5c08f6738ade","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1508.2299700220494,"rawValue":1508.2299700220494,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=427","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-ce242a39a3d06be5babaae48","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1475.3522024096037,"rawValue":1475.3522024096037,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=804","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-43ed5188543944ce29893cc9","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1482.767837213196,"rawValue":1482.767837213196,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=796","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c0f62f3c1c154b11ce4d9ddb","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1440.7741507579944,"rawValue":1440.7741507579944,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=60","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-a13a6d7b45dc0cbb722be80a","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1545.239658145461,"rawValue":1545.239658145461,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=399","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-be4641c77c983cc0fea1e314","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1481.889534951889,"rawValue":1481.889534951889,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=10","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-3b8e4361498015d8ed0d14cd","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-text","benchmarkName":"Arena Text","benchmarkVersionHint":"arena-text@latest","metricId":"arena_score_bt","value":1499.1631125049637,"rawValue":1499.1631125049637,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=771","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"text","category":"coding","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-fffcaaa4edf3494b9bf5f642","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1366.0454732568385,"rawValue":1366.0454732568385,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=409","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c52ac83e98f4be3a487e477f","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1328.0663081312318,"rawValue":1328.0663081312318,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=0","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-bd2cb0398b9ad05ee08ab84b","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1356.2812075554702,"rawValue":1356.2812075554702,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=706","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-23060da405f4f69623b3d036","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1329.250814778743,"rawValue":1329.250814778743,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=512","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-02bb5d29d8a00bab3f09afce","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1339.4591306887673,"rawValue":1339.4591306887673,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=884","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-2277d8e9bd5e881a93d68473","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1339.609797891108,"rawValue":1339.609797891108,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=324","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-3724a34e8fe3910ac14115a8","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1330.367604811887,"rawValue":1330.367604811887,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=806","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-ef7ef9de0eff99fb5c1ce602","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1386.2439096452076,"rawValue":1386.2439096452076,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=182","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-8522d0d383fbaaa00a5689f5","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1253.4276980479322,"rawValue":1253.4276980479322,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=297","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-01-09","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-55a9268e34a69a244341db36","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1224.2350785852213,"rawValue":1224.2350785852213,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=777","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-aecdc1d5125754b8a43209fd","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1178.450043786878,"rawValue":1178.450043786878,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=967","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-b2674af7af509454f9599368","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1175.5783200444741,"rawValue":1175.5783200444741,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=494","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-8a5840fbc64250c20cd2b1ab","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1186.696775338311,"rawValue":1186.696775338311,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=597","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-08c3ade848d7a4bcee3d4675","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1175.4214503141313,"rawValue":1175.4214503141313,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=87","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-515bbf66032a23291646b27d","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514-thinking-16k","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1195.870868422259,"rawValue":1195.870868422259,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=960","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-362309543a4e5781d922676a","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514-thinking-16k","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1201.0411343597555,"rawValue":1201.0411343597555,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=784","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-9d0dd3210e541cd607e74e90","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514-thinking-16k","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1191.1950598829105,"rawValue":1191.1950598829105,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=79","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-fc7a0d7a6cfd8f5f1ab3b310","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514-thinking-16k","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1175.315091204888,"rawValue":1175.315091204888,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=495","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-0f30aec3cb97a88d33e0687a","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514-thinking-16k","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1183.6309405267036,"rawValue":1183.6309405267036,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=599","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-ac91dce894e08737886f2069","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1339.6258946871856,"rawValue":1339.6258946871856,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=411","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c050e3b10f609c664768d342","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1319.34049373008,"rawValue":1319.34049373008,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=808","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-91f23ca3d2e46a109854aa64","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1325.267920995656,"rawValue":1325.267920995656,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=888","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-26981c01b893a195e4006fa4","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1380.6075943954047,"rawValue":1380.6075943954047,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=183","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-e57a2313ac40c64c4d008488","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1315.8592803719687,"rawValue":1315.8592803719687,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=517","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-20718926b0347f1acf726c73","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1316.7408184183773,"rawValue":1316.7408184183773,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=329","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":12,"sourceMetricId":"elo"},{"id":"pub-f33d570ea7b711ae1aabac3b","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1311.3488129222949,"rawValue":1311.3488129222949,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=7","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":13,"sourceMetricId":"elo"},{"id":"pub-516f530a84f8def022ed1943","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1328.281760751524,"rawValue":1328.281760751524,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=718","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":17,"sourceMetricId":"elo"},{"id":"pub-479ebf6561042b93e3574158","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1254.8661093976898,"rawValue":1254.8661093976898,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=675","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":18,"sourceMetricId":"elo"},{"id":"pub-24147305a0adfe747520b423","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1331.6823510071824,"rawValue":1331.6823510071824,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=325","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-fb2f05f2669069bae4ab6810","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1325.798127104686,"rawValue":1325.798127104686,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=887","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-212acbdbdbea960b7dd24c2d","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1319.3019854373526,"rawValue":1319.3019854373526,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=515","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-9861f7a02fb8660c0826efa6","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1314.5661177451846,"rawValue":1314.5661177451846,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=5","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-c75ea9f46c1a43407bb0eb62","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1308.616440051339,"rawValue":1308.616440051339,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=810","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-654413bc529fe7f897ee313e","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1331.2485350549919,"rawValue":1331.2485350549919,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=415","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-4d515c8cc99febee54875edf","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1372.6502759169039,"rawValue":1372.6502759169039,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=186","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":14,"sourceMetricId":"elo"},{"id":"pub-14cc2df196584d7e9c5d6efe","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1335.4145440824989,"rawValue":1335.4145440824989,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=714","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":15,"sourceMetricId":"elo"},{"id":"pub-60837682f69296def843425b","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1277.7696033143948,"rawValue":1277.7696033143948,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=668","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":16,"sourceMetricId":"elo"},{"id":"pub-da8386e63dde5ce1db6a3772","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1321.4273346920506,"rawValue":1321.4273346920506,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=513","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-d507c885400b90ab6dd7a046","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1316.5756460386099,"rawValue":1316.5756460386099,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=2","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-838ba333508b52f1ef802d32","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1325.846767304178,"rawValue":1325.846767304178,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=886","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-ebc0aabe2073286ae57ca0ef","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1336.5227447457085,"rawValue":1336.5227447457085,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=413","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-0d0d366aa7fcaa2022bffe35","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1336.323477600447,"rawValue":1336.323477600447,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=712","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":12,"sourceMetricId":"elo"},{"id":"pub-e92792b0b56ddf43c3c9b7fe","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1311.0181433353632,"rawValue":1311.0181433353632,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=331","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":13,"sourceMetricId":"elo"},{"id":"pub-a790974467dbb738b0f37d1a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1368.6043696266045,"rawValue":1368.6043696266045,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=187","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":14,"sourceMetricId":"elo"},{"id":"pub-6ccb68513e7d60f246bab263","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1303.2734404857263,"rawValue":1303.2734404857263,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=813","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":15,"sourceMetricId":"elo"},{"id":"pub-163694879faef97c3353b6da","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1319.8858391023387,"rawValue":1319.8858391023387,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=514","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-7f33c292a76091100ba734a2","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1315.5218357271092,"rawValue":1315.5218357271092,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=3","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-380c9fe8040247213057bd98","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1338.363607600836,"rawValue":1338.363607600836,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=412","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-9124d2ea1240baae49da93dc","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1323.0092205490803,"rawValue":1323.0092205490803,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=327","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-f252ae39a7e4fc2a2ea5d501","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1309.4296577335394,"rawValue":1309.4296577335394,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=809","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-f09196ca8308e00a5a3d9a45","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1337.4369448602456,"rawValue":1337.4369448602456,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=711","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-8294ce121ad98d212fb0d8d5","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1322.8715236900614,"rawValue":1322.8715236900614,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=890","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-7a6c3bcdb8d8f4b91ada46c8","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1358.969335041679,"rawValue":1358.969335041679,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=190","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":16,"sourceMetricId":"elo"},{"id":"pub-00ce8e509765e4db55ad5050","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1263.3991203690257,"rawValue":1263.3991203690257,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=671","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":17,"sourceMetricId":"elo"},{"id":"pub-228c043b7f183ffd7696eb24","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1327.2926742680615,"rawValue":1327.2926742680615,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=719","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-1320625131a7853a3b559217","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1291.734806271278,"rawValue":1291.734806271278,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=529","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-a981d5d6479d425dd96e1fc4","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1346.5824304077719,"rawValue":1346.5824304077719,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=198","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-7b330ac1088f7a19e5c30573","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1287.2702102794583,"rawValue":1287.2702102794583,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=21","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":12,"sourceMetricId":"elo"},{"id":"pub-6f02408b44df2c46d07be672","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1298.9715749528548,"rawValue":1298.9715749528548,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=905","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":13,"sourceMetricId":"elo"},{"id":"pub-2e929dfdc3eb0e8a6154d0fe","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1305.244908350073,"rawValue":1305.244908350073,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=430","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":14,"sourceMetricId":"elo"},{"id":"pub-800b9c3e5c30edeedbac1149","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1274.6554218727329,"rawValue":1274.6554218727329,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=345","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":15,"sourceMetricId":"elo"},{"id":"pub-904825acd47ce770e2cddbe0","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1266.5761668794835,"rawValue":1266.5761668794835,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=836","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":16,"sourceMetricId":"elo"},{"id":"pub-2ac4ddf1d110ef1d2bd85cc7","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1340.212438247605,"rawValue":1340.212438247605,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=709","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-104db0ff1febd3de578f58e9","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1307.675216355538,"rawValue":1307.675216355538,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=332","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-519e1749e42eccfbd01e46a8","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1291.6511377416784,"rawValue":1291.6511377416784,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=815","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-27b89d99f429397f5ce8156b","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1317.0928207349784,"rawValue":1317.0928207349784,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=422","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-0f11a15cf5704b25f9f38f70","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1305.235014637116,"rawValue":1305.235014637116,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=898","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-ecc5097f7115541bfb4ab6a9","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1293.284087685157,"rawValue":1293.284087685157,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=17","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-b690e32c00b76108416891d0","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1291.0826155661205,"rawValue":1291.0826155661205,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=530","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-53a6ee166fe47b24b1bda8cb","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1344.3851109643263,"rawValue":1344.3851109643263,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=199","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-eb9a7f1e4d244b382284cac0","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1409.7286304046831,"rawValue":1409.7286304046831,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=179","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-7d9a2a5ffe328c137b61b5b3","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1348.1123538602462,"rawValue":1348.1123538602462,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=410","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-1ede6cff42245f6a3ef0a412","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1342.11414493275,"rawValue":1342.11414493275,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=707","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-0b3f953b891a6afd4e3b1ff9","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1333.0919622997974,"rawValue":1333.0919622997974,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=885","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-031539d43962f0fc3df65acf","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1322.8117958326172,"rawValue":1322.8117958326172,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=1","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-e05754b1465433f3c945018c","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1318.806514049852,"rawValue":1318.806514049852,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=516","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-d276a39cb10f7a506ce2c6ed","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1294.7404400117455,"rawValue":1294.7404400117455,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=336","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-0d297e3a09334b7f4ca7043a","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1288.9859217042076,"rawValue":1288.9859217042076,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=817","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-bf7f2059fbebee177049669e","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1296.829101074667,"rawValue":1296.829101074667,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=524","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-5426afa5e14c8d9cd0f07c80","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1305.0716357393028,"rawValue":1305.0716357393028,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=431","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-1eb75aa20595009227acc3b7","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1310.6321671807964,"rawValue":1310.6321671807964,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=728","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-e4f5fa13ae6d303b5d5e79fe","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1295.4393376689159,"rawValue":1295.4393376689159,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=907","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-6fbabf9fba70a378f23231b5","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1281.4891028508862,"rawValue":1281.4891028508862,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=23","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-5ae72831e758f4a674f1eb12","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1265.842434488888,"rawValue":1265.842434488888,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=352","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-ee8b6076380004989293815f","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1309.6732366965077,"rawValue":1309.6732366965077,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=214","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-a6fd86de0ea2edd0d137d3b9","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1188.9836903273174,"rawValue":1188.9836903273174,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=696","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-d79395e4ee36c94599bf65d2","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1239.3029902712422,"rawValue":1239.3029902712422,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=846","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-a5990b69aa82910ea5c41afd","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1322.0258604166127,"rawValue":1322.0258604166127,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=721","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-ecbdb68c8d03a2e4219c0e2a","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1310.9250368004123,"rawValue":1310.9250368004123,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=426","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-d2bb56a9cf8b8c98cfaa7bad","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1296.412961143429,"rawValue":1296.412961143429,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=906","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-d0dfffae6a102685ae6af6f6","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1279.2597667873204,"rawValue":1279.2597667873204,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=826","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-3330fed95b1db2401c5ee7e3","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1272.5797654897196,"rawValue":1272.5797654897196,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=346","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-2dc5d7db938426ff5b9fed05","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1338.9932788275976,"rawValue":1338.9932788275976,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=202","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-4ed09589ed46364d26112663","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1281.1515471513026,"rawValue":1281.1515471513026,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=24","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-60d33b0c20e81da9526fdd04","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1278.225200016402,"rawValue":1278.225200016402,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=537","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-074ebd8e0c807b7b377cdcb3","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1277.975981927435,"rawValue":1277.975981927435,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=152","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"captioning","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-f315ab53cf8b729fd54625cc","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1253.054839283779,"rawValue":1253.054839283779,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=298","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-01-09","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-efe24e630bef37f750338e57","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1250.9360691466372,"rawValue":1250.9360691466372,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=676","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-0a4616aa3b956f7a5c33b9b2","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1285.4962061183683,"rawValue":1285.4962061183683,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=225","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-08ce87c24f33f98ad653dbfc","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1245.0960097197624,"rawValue":1245.0960097197624,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=372","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-7782ebdcbc5d06f09a674097","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1223.3192822779315,"rawValue":1223.3192822779315,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=855","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-d828f930e1b63e1e9d7d8ebb","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1241.5039645417346,"rawValue":1241.5039645417346,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=568","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-d7fa455adfc1bd7a88cb1f13","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1235.522601905443,"rawValue":1235.522601905443,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=57","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-8f443362e4717a5b7c1facfa","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1244.7952459440392,"rawValue":1244.7952459440392,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=766","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-04f77d0cbf256854f7958c04","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1246.40573018332,"rawValue":1246.40573018332,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=470","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-209dd98a22aa16dff3ad06e7","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1238.3239190626143,"rawValue":1238.3239190626143,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=946","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-f26aa28fd94cb079925a61b9","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1302.419804626094,"rawValue":1302.419804626094,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=149","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"captioning","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-23886caaa51b3fb81a444754","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1277.3617983725967,"rawValue":1277.3617983725967,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=295","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-01-09","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-79a10af4d58166d167a12736","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1283.3072105961328,"rawValue":1283.3072105961328,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=667","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-1b75c2112bf7e236c8298949","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1269.9078347737168,"rawValue":1269.9078347737168,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=349","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-cbfa8b08b0836cf6bd9e2149","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1261.6808522534452,"rawValue":1261.6808522534452,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=838","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-dd2e6ed2adf313c5617f9cc5","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1280.950443411543,"rawValue":1280.950443411543,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=743","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-0f559e37c61702543fc66a48","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1269.4436691427518,"rawValue":1269.4436691427518,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=921","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-deb789af9cb5df83c25520a6","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1262.0089168132201,"rawValue":1262.0089168132201,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=39","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-fa452b7e6b5542892300bffa","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1271.765841763226,"rawValue":1271.765841763226,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=449","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-d506f997a38b9fcf90ed5e9f","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1261.9961975859537,"rawValue":1261.9961975859537,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=552","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-27a3abfc3bf50da46d1223fb","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1292.1401419513602,"rawValue":1292.1401419513602,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=223","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-c3be4934cea16eb8a1fc5a96","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1335.876900901611,"rawValue":1335.876900901611,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=147","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"captioning","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-cac301324492df6b72d3a845","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1327.4357508689236,"rawValue":1327.4357508689236,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=661","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-79fba7f6e50be8dae49e82f0","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1325.9836234703632,"rawValue":1325.9836234703632,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=807","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-42387ebb6e7ff8596053867b","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1322.3070079699012,"rawValue":1322.3070079699012,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=328","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-6a8d27e4e05c0d144b426486","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1367.215332507108,"rawValue":1367.215332507108,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=188","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-9b00f8bc7048ac0226d84d96","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1312.6024874736008,"rawValue":1312.6024874736008,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=425","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-5aa079b81fb0f997c741aaf2","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1293.9433085543112,"rawValue":1293.9433085543112,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=16","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-0534bd22954965866dab79fc","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1316.8798165983414,"rawValue":1316.8798165983414,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=722","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-1a297e86a1666a3826898131","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1303.4830818894375,"rawValue":1303.4830818894375,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=901","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-4c890fce89d1d0879b53aaa0","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1287.058712057561,"rawValue":1287.058712057561,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=534","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-a219db01d9749425059343be","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1387.6642022682222,"rawValue":1387.6642022682222,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=180","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-a10e3501fa8470e24939dabb","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1312.7261642378835,"rawValue":1312.7261642378835,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=6","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-ec994bfc0a54993b082d89f9","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1304.679722098694,"rawValue":1304.679722098694,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=519","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-9349f635ad04ba99c9e89ffb","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1308.561229527967,"rawValue":1308.561229527967,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=811","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-f408e89994beab7d232e23bc","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1314.1394722686966,"rawValue":1314.1394722686966,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=893","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-6c19933d1c9a065281a383c1","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1320.13295768415,"rawValue":1320.13295768415,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=418","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-71ed0a759cb876e72a223299","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1290.5413180022827,"rawValue":1290.5413180022827,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=337","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":15,"sourceMetricId":"elo"},{"id":"pub-2f0f4726516108bb2f1e8591","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1308.5422569432683,"rawValue":1308.5422569432683,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=730","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":16,"sourceMetricId":"elo"},{"id":"pub-418939c58ef5c638ef014f7a","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1297.5949460560837,"rawValue":1297.5949460560837,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=434","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-cf38bb216a00972ca497f9e2","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1284.3176000631856,"rawValue":1284.3176000631856,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=913","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-db6b5580db1a87c27ad05639","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1272.4891253469764,"rawValue":1272.4891253469764,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=31","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-905d70948e954bc03be4bb07","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1264.951758641474,"rawValue":1264.951758641474,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=548","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-5dfa44709076479b5d8093da","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1282.3434805213312,"rawValue":1282.3434805213312,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=742","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-af136a2a6e511d48cc652ba7","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1249.7779351419053,"rawValue":1249.7779351419053,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=365","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-31091102d5183c84679a611d","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1387.5118900481539,"rawValue":1387.5118900481539,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=181","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-11cf8960dd7c2b5a3dedabda","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1315.7027920506393,"rawValue":1315.7027920506393,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=891","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-940e17d23862163fac63cb59","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1303.7316413327524,"rawValue":1303.7316413327524,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=812","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-3ac0fc9cc0289bf4cbde1f60","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1308.2050350788284,"rawValue":1308.2050350788284,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=8","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-ff320d63f128bee8b1c7c448","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1300.950541557701,"rawValue":1300.950541557701,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=522","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-d9794a6cdf5821d2944b1c05","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1330.236284451325,"rawValue":1330.236284451325,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=716","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":12,"sourceMetricId":"elo"},{"id":"pub-0029e2ee791ec6e8ae1aa762","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1303.98819623092,"rawValue":1303.98819623092,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=333","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":13,"sourceMetricId":"elo"},{"id":"pub-b474d8ff8f508c518c06b73f","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1317.055515893644,"rawValue":1317.055515893644,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=423","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":14,"sourceMetricId":"elo"},{"id":"pub-cbe404e3c58c901ab4e923bd","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1340.9007264618651,"rawValue":1340.9007264618651,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=708","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-850c1b2be9e1262d766b7729","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1314.7896890894135,"rawValue":1314.7896890894135,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=892","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-e2d1f3667c7cef3708ce67e3","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1296.9188367436175,"rawValue":1296.9188367436175,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=523","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-185b526449c1e84f163183b1","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1301.9064394211337,"rawValue":1301.9064394211337,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=12","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-52f9907ecb3c0bf8ecc683c2","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1285.890335789169,"rawValue":1285.890335789169,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=341","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-3f79066b5e8387055fd45ef8","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1310.3261723682426,"rawValue":1310.3261723682426,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=428","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-85fb984588a14e12527033a9","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1268.647085741865,"rawValue":1268.647085741865,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=156","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"captioning","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-bbf14a34c6dca2b81bb1127a","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1244.1771380630535,"rawValue":1244.1771380630535,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=678","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-12c4dbe80cfdc99b35254537","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1275.2358717078994,"rawValue":1275.2358717078994,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=829","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-c2d5d14e076eec967c41e374","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1274.897115188257,"rawValue":1274.897115188257,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=538","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-48029c0af66df4e8a46bc8ee","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1330.5232576900614,"rawValue":1330.5232576900614,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=205","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-c76e00033ae42f42dc1898e7","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1281.3861990362748,"rawValue":1281.3861990362748,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=443","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-3418f1e4efa8c79bb7e55c97","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1261.994052083512,"rawValue":1261.994052083512,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=357","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-029204fa679bab5d7bba279d","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1263.5451694367357,"rawValue":1263.5451694367357,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=37","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-71a6176400e29d8d8678862d","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1268.5293700175055,"rawValue":1268.5293700175055,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=922","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-f3d00622ab34cc02f96afffc","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1273.281791266762,"rawValue":1273.281791266762,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=750","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-dac809a9ec7c6d039a031fcb","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1259.6186794427995,"rawValue":1259.6186794427995,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=672","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-b2e4e9110ad6a2298a1b2ba1","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1201.9932948782132,"rawValue":1201.9932948782132,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=167","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"captioning","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-9637763ed9058419f47f1c6f","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1188.8295591249853,"rawValue":1188.8295591249853,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=309","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-01-09","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-f348e587cbebc73c69b28a74","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1210.0323470503438,"rawValue":1210.0323470503438,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=861","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-880facd7a70fee5b29797f71","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1239.457444566454,"rawValue":1239.457444566454,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=240","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-f82aea71d7fd69a2d7763607","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1201.2199259431375,"rawValue":1201.2199259431375,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=390","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-c658d37efe83d5de50e66fbd","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1228.9053106371352,"rawValue":1228.9053106371352,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=476","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-d2b4f539ad26eca35891894d","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1229.885425732703,"rawValue":1229.885425732703,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=774","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-ccc72f09dc82a056a6956a54","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1211.316408367025,"rawValue":1211.316408367025,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=68","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-ba52fe43745a2ed7b891f643","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1217.908476418474,"rawValue":1217.908476418474,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=952","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-611eed9732417e21f2decd75","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1220.0929625863241,"rawValue":1220.0929625863241,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=580","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-2c9bacc11e087e3dbd291b89","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1213.799181897436,"rawValue":1213.799181897436,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=163","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"captioning","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-f9ecba5fa592bde07c7afed0","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1223.6468529366466,"rawValue":1223.6468529366466,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=684","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-4f6a870b82131185589d7c8b","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1181.3109549238002,"rawValue":1181.3109549238002,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=314","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-01-09","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-0b27c3eb6a4f90acf909f3ea","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1206.7717125175222,"rawValue":1206.7717125175222,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=862","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-5a30a5857d1bd5af86d5faf4","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1216.1474920472922,"rawValue":1216.1474920472922,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=248","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-d6d41dcb0afbb233424646b0","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1216.2514858098239,"rawValue":1216.2514858098239,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=583","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-a1193a3be4055a4d68a269d3","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1192.9181362205732,"rawValue":1192.9181362205732,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=394","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-3a7708c8ec8397a204b21457","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1202.7394433197264,"rawValue":1202.7394433197264,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=72","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-d02268f7769300b1a0fa04c3","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1204.0576519797155,"rawValue":1204.0576519797155,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=482","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-84b542ed814653a70cbe3c19","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1199.8221347399667,"rawValue":1199.8221347399667,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=957","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-57bad14cb2ed223a79567641","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1208.2126541809084,"rawValue":1208.2126541809084,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=782","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-b5221e5de3ed7881e6c60b2f","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1117.231591275592,"rawValue":1117.231591275592,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=321","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-01-09","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-6c2eb9dc1fa202c159366fe2","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1112.4783088166164,"rawValue":1112.4783088166164,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=881","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-502f49172157e34f69ca4365","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1160.112291219351,"rawValue":1160.112291219351,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=797","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-c582ba44f4a245b4cf46732d","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1154.7895306986513,"rawValue":1154.7895306986513,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=976","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-b8f210aa0842cee4ee742758","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1173.274753015647,"rawValue":1173.274753015647,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=605","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-9d1000cc947329649a10b3d3","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1162.2950013064317,"rawValue":1162.2950013064317,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=93","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-ca51bc5609aa60cec99596a0","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1140.2694032439917,"rawValue":1140.2694032439917,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=503","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-2f6d6acd56e292ddf4f4dab2","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1291.2643112763622,"rawValue":1291.2643112763622,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=292","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-01-09","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-ec6a0acc7c62d59c203d18e4","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1182.2640435663832,"rawValue":1182.2640435663832,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=170","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"captioning","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-2f45cc22bab7d48619c25557","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1254.939859587545,"rawValue":1254.939859587545,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=361","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-7956a3b258962bda18c07dd0","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1245.0680842058491,"rawValue":1245.0680842058491,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=845","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-4fa33b15b6ec678486726e0d","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1183.2305966439017,"rawValue":1183.2305966439017,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=700","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-a22a425768287c038840504d","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1261.8111902855208,"rawValue":1261.8111902855208,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=236","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":17,"sourceMetricId":"elo"},{"id":"pub-7c19b93fb044010b7b8c5323","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1242.8759948131935,"rawValue":1242.8759948131935,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=942","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":18,"sourceMetricId":"elo"},{"id":"pub-4554be4d1ce594ebd6a887e4","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1232.9799064050783,"rawValue":1232.9799064050783,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=59","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":19,"sourceMetricId":"elo"},{"id":"pub-4f71df717dd635712085dcf6","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1242.5364839891727,"rawValue":1242.5364839891727,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=767","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":20,"sourceMetricId":"elo"},{"id":"pub-ff4100cd89c55e30eac9a872","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1245.8305564779262,"rawValue":1245.8305564779262,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=472","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":21,"sourceMetricId":"elo"},{"id":"pub-ad28b39166e34cc91dbb29b5","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1231.4428551515907,"rawValue":1231.4428551515907,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=575","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":22,"sourceMetricId":"elo"},{"id":"pub-0071934ea056f5428cebd381","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1282.407597850386,"rawValue":1282.407597850386,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=294","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-01-09","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-059255f894d39ee4963ea32b","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1243.2963597355006,"rawValue":1243.2963597355006,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=158","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"captioning","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-4cde87c15da2c6431d907373","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1242.6841746921289,"rawValue":1242.6841746921289,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=679","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-536009653a355aa11a11fd27","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1253.3957374079444,"rawValue":1253.3957374079444,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=363","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-b2a4b99cef017c49a836953c","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1272.0415127715962,"rawValue":1272.0415127715962,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=751","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-ece0b8a396103edafe3719ee","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1234.2638406611563,"rawValue":1234.2638406611563,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=850","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-bea2852c52f079998214a9c3","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1248.6543411367695,"rawValue":1248.6543411367695,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=47","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":12,"sourceMetricId":"elo"},{"id":"pub-728b381ce09977deff6d1c5c","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1252.3508828972876,"rawValue":1252.3508828972876,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=560","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":13,"sourceMetricId":"elo"},{"id":"pub-5c91957f843cbb938913cb06","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1262.3312416657832,"rawValue":1262.3312416657832,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=457","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":14,"sourceMetricId":"elo"},{"id":"pub-c5409ab61bfb41980f3b18bc","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1279.0259454710808,"rawValue":1279.0259454710808,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=229","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":15,"sourceMetricId":"elo"},{"id":"pub-dd45dc96fcbf9b0d0a166807","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1252.1692312332254,"rawValue":1252.1692312332254,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=935","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":16,"sourceMetricId":"elo"},{"id":"pub-4188bfce9c4362c522d3d5e2","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1166.9589829971826,"rawValue":1166.9589829971826,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=172","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"captioning","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-7ddef859f7fe169f15c398cd","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1207.3004207354725,"rawValue":1207.3004207354725,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=690","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-ffb042177c4fa74306d82d3b","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1261.9939078972589,"rawValue":1261.9939078972589,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=757","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-a17ce979fcd81f50a51e8c75","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1270.1309312878475,"rawValue":1270.1309312878475,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=231","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":12,"sourceMetricId":"elo"},{"id":"pub-e17c57baea80fad92b2b8e11","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1223.0603952349345,"rawValue":1223.0603952349345,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=383","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":15,"sourceMetricId":"elo"},{"id":"pub-8df1d56c155e127b56f30579","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1229.5374284749435,"rawValue":1229.5374284749435,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=61","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":16,"sourceMetricId":"elo"},{"id":"pub-9a7517cdd6184042720d01ef","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1246.018575021936,"rawValue":1246.018575021936,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=471","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":17,"sourceMetricId":"elo"},{"id":"pub-faf41e37273cc408b171b2f7","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1229.8209084385103,"rawValue":1229.8209084385103,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=576","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":18,"sourceMetricId":"elo"},{"id":"pub-056e82409d8efcc5aa9e4c56","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1192.119362312469,"rawValue":1192.119362312469,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=868","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":19,"sourceMetricId":"elo"},{"id":"pub-66a569c4e65b49da08042960","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1234.0043090485344,"rawValue":1234.0043090485344,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=949","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":20,"sourceMetricId":"elo"},{"id":"pub-6cf6e000fb430f7e8a68ce8f","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1275.334826947226,"rawValue":1275.334826947226,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=153","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"captioning","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-f399bfc585751c1d9a2a6a26","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1212.7954404620546,"rawValue":1212.7954404620546,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=689","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-820c6fefca64950d8dcad030","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1279.155085162679,"rawValue":1279.155085162679,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=746","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-dfb1c588fba125618eb730d8","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1294.9368449319488,"rawValue":1294.9368449319488,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=221","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-0998329398afc655df25e1b0","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1263.948084493556,"rawValue":1263.948084493556,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=455","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-0193c479f46fb56eee36f62c","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1256.455683799964,"rawValue":1256.455683799964,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=930","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-5130a9b3ab0b38a7e7e3a342","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1229.8091767270755,"rawValue":1229.8091767270755,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=852","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-abe365af02a6ad50c06cfb0d","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1246.606547156016,"rawValue":1246.606547156016,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=51","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-9598bde15f78b9079bcb8361","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1244.7602693452304,"rawValue":1244.7602693452304,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=565","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":13,"sourceMetricId":"elo"},{"id":"pub-4f8069979cfe2b1638467719","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1224.2536232112043,"rawValue":1224.2536232112043,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=380","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":14,"sourceMetricId":"elo"},{"id":"pub-2b0bd34cc069cc4f4aa06aa7","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1318.6922541052932,"rawValue":1318.6922541052932,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=421","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-77db52d0443573b735f23c15","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1258.7874558529852,"rawValue":1258.7874558529852,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=673","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-edc0e05970769a92ed8dec03","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1292.7790098586884,"rawValue":1292.7790098586884,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=19","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-3439e518da4b7310e509b92c","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1300.6105320537492,"rawValue":1300.6105320537492,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=903","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-31f74f6392ca873bf0bbb514","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1288.935355180308,"rawValue":1288.935355180308,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=532","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":12,"sourceMetricId":"elo"},{"id":"pub-417db59194ddbd6ab7cd686a","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1311.6533468830087,"rawValue":1311.6533468830087,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=727","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":13,"sourceMetricId":"elo"},{"id":"pub-e3e4a5ea671adc17407ca8f2","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1341.416188386232,"rawValue":1341.416188386232,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=201","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":14,"sourceMetricId":"elo"},{"id":"pub-d7eb41a010992c7b22ee0e1d","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1266.977114552296,"rawValue":1266.977114552296,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=351","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":16,"sourceMetricId":"elo"},{"id":"pub-488e204093d5d8e0317a64ae","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1257.4700485614374,"rawValue":1257.4700485614374,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=842","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":18,"sourceMetricId":"elo"},{"id":"pub-3c1b3d6c9672a4f9c004beab","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1322.4451062321143,"rawValue":1322.4451062321143,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=417","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-46168d81586f916e3514ae78","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1335.0946163092026,"rawValue":1335.0946163092026,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=715","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-1d06f13b548d5e51181b85e4","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1307.2918403062465,"rawValue":1307.2918403062465,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=896","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-067b3ef6824687bbecbda7af","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1357.0649498594507,"rawValue":1357.0649498594507,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=191","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-e8baefadf41ce4099f8e3028","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1297.0944905473443,"rawValue":1297.0944905473443,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=13","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-dbfe7ea6483d9484c7406687","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1295.03599253479,"rawValue":1295.03599253479,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=525","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-e026dcd7669aae1d367db7b9","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1283.3210953505347,"rawValue":1283.3210953505347,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=822","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-309b171b7ebbe7edac09bdf8","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1271.5249479732902,"rawValue":1271.5249479732902,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=347","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":15,"sourceMetricId":"elo"},{"id":"pub-959caa332feab6d82c18f626","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1217.7027405834353,"rawValue":1217.7027405834353,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=687","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":17,"sourceMetricId":"elo"},{"id":"pub-671a1b9eb78fc84328af5aca","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1287.2696508457348,"rawValue":1287.2696508457348,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=737","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-64d0ae50f999964df2016f32","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1203.5147616249446,"rawValue":1203.5147616249446,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=692","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-c1e20f1168690f279cac3239","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1293.5350001134723,"rawValue":1293.5350001134723,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=222","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-5d71f85b9e8843796500881c","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1235.6731271897654,"rawValue":1235.6731271897654,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=847","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-bd171678a4b1e52b8550afaf","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1264.299027786259,"rawValue":1264.299027786259,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=454","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-a7bfd920b2423ce06177b4a9","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1252.9412328638289,"rawValue":1252.9412328638289,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=934","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-d3cbc373d71980cd989b2981","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1245.177396795918,"rawValue":1245.177396795918,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=53","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-736ef97a277f93473659619c","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1242.662480228433,"rawValue":1242.662480228433,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=567","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-17796ac6cc61d2fa49b57c6a","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1220.2872792396493,"rawValue":1220.2872792396493,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=385","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-fa31dde01cc41315a76baef0","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1142.9404226060835,"rawValue":1142.9404226060835,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=704","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-4440457b63a1e63e69c68126","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1233.9470317021821,"rawValue":1233.9470317021821,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=241","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-f1ecbb4207e1ad214990198a","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1226.0625736431061,"rawValue":1226.0625736431061,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=776","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-a92f355fc7692316d95efe7e","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1211.8757458712262,"rawValue":1211.8757458712262,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=480","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-80fa159e76d715d373e087d0","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1206.07617573138,"rawValue":1206.07617573138,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=955","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-192e69837b55c718d16bba30","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1164.3312245550119,"rawValue":1164.3312245550119,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=876","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-d63e921a2e4538c10cf63bce","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1197.7172341515723,"rawValue":1197.7172341515723,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=75","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-baa71d486f6d1546f06acc9c","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1200.6826419414947,"rawValue":1200.6826419414947,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=588","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-ed9eeafa2c41051c6d4ea6dc","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1165.204779654008,"rawValue":1165.204779654008,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=402","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-b1ce5b848c9bdc8af196cde0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1323.7962585878583,"rawValue":1323.7962585878583,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=416","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-322a147d5646ddeb30b8f170","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1329.394408300886,"rawValue":1329.394408300886,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=717","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-73975d060a229cc6164fe00e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1293.580997071454,"rawValue":1293.580997071454,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=526","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-50d8170d6c3c30a73d1070e4","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1295.3180242128233,"rawValue":1295.3180242128233,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=14","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-ff7e7ee9a6e50af50c3a1ab3","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1304.866839783428,"rawValue":1304.866839783428,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=900","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-ee546d8c91a4d0fd6d49e6cd","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1351.9563378544017,"rawValue":1351.9563378544017,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=196","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":12,"sourceMetricId":"elo"},{"id":"pub-278b2b4417f8fec4c2aec886","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1276.8317204106327,"rawValue":1276.8317204106327,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=344","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":15,"sourceMetricId":"elo"},{"id":"pub-2d2dd8eb98a0d25412dd0e3c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1278.8881399064164,"rawValue":1278.8881399064164,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=827","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":16,"sourceMetricId":"elo"},{"id":"pub-29c6e49f3c6750b4418b859f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1336.2833885788102,"rawValue":1336.2833885788102,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=713","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-ff1e3756dc88f9eeaf91bf3f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1298.8071058294759,"rawValue":1298.8071058294759,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=334","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-0177788e4a7a34f66f4ed8bc","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1291.514628420573,"rawValue":1291.514628420573,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=816","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-0a0b6d2edfd6faa6a1b43bea","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1356.4797860041688,"rawValue":1356.4797860041688,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=192","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-895d5a0fb29ba883140450ed","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1304.9281179691434,"rawValue":1304.9281179691434,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=899","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-4185b37e839c44a7cb784dad","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1315.5795724452862,"rawValue":1315.5795724452862,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=424","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-3b1fbe982e3100157c01e108","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1293.1376867171932,"rawValue":1293.1376867171932,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=18","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":13,"sourceMetricId":"elo"},{"id":"pub-dfef7eca756dab6f0dcd61aa","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1290.5116187642604,"rawValue":1290.5116187642604,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=531","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":14,"sourceMetricId":"elo"},{"id":"pub-74f96fe001dfad16e1d020c6","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1270.842498626393,"rawValue":1270.842498626393,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=832","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":17,"sourceMetricId":"elo"},{"id":"pub-d08ab9e0051e1b90930d45f9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1275.86542240388,"rawValue":1275.86542240388,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=447","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":18,"sourceMetricId":"elo"},{"id":"pub-30588b82f748c92b61564c07","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1250.6990774461399,"rawValue":1250.6990774461399,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=45","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":19,"sourceMetricId":"elo"},{"id":"pub-696e4b615ef78f66971ab6e0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1254.571315524965,"rawValue":1254.571315524965,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=558","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":20,"sourceMetricId":"elo"},{"id":"pub-0139a1cb9f0c395dee79812e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1247.6822220510298,"rawValue":1247.6822220510298,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=369","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":21,"sourceMetricId":"elo"},{"id":"pub-abe70bc2f3101714142813a3","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1256.3214133691158,"rawValue":1256.3214133691158,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=931","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":22,"sourceMetricId":"elo"},{"id":"pub-2a8b4c693168601984db4814","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1270.1397680387188,"rawValue":1270.1397680387188,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=230","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":23,"sourceMetricId":"elo"},{"id":"pub-8ef1dbe2e0ce977be1dbcc04","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1252.7929433303175,"rawValue":1252.7929433303175,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=763","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":24,"sourceMetricId":"elo"},{"id":"pub-dc61d907026163ea04dbfba5","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1283.9878695740285,"rawValue":1283.9878695740285,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=741","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-d7549212e9298135c0408ed6","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1268.806707422959,"rawValue":1268.806707422959,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=450","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-6dd7fbd10b1dfe31075f1bad","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1297.0964190188013,"rawValue":1297.0964190188013,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=220","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-fb96fe1e013e385322154850","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1258.0785392330015,"rawValue":1258.0785392330015,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=555","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-333bab6e005a2122c2a400e9","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1248.502963204978,"rawValue":1248.502963204978,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=366","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-2e5035abd9663d21f807b769","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1253.5102872507298,"rawValue":1253.5102872507298,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=44","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-d4b0f34b596284cabf11e4f6","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1256.5801748871338,"rawValue":1256.5801748871338,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=929","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-34f7638fd7582fe7f2cc565e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1214.6377754173036,"rawValue":1214.6377754173036,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=860","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-2897c0b2589b6f62232064e0","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1316.23932965299,"rawValue":1316.23932965299,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=723","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-6205a70e56c7a64a24fd1b67","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1297.3044350302314,"rawValue":1297.3044350302314,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=435","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-c124c36e6b991825441b0099","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1285.4465654380217,"rawValue":1285.4465654380217,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=911","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-70ed4ef5f2e56f35ed746a54","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1275.4883658285985,"rawValue":1275.4883658285985,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=28","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-0b85dd1a4b52f159d8617ecf","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1274.090876532495,"rawValue":1274.090876532495,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=540","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-a3c2f08f971f8e85f3e58821","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1261.1758740188739,"rawValue":1261.1758740188739,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=358","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-19b0c53ead2abe1261b2a9bf","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1258.359664515613,"rawValue":1258.359664515613,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=841","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-a8e7870fb5822b414eae77ba","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1297.6844597348656,"rawValue":1297.6844597348656,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=219","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-eaedccea7035fb85151e334c","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1374.2578449041494,"rawValue":1374.2578449041494,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=185","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-e2ce6f6928e25cc08f9d844b","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1316.024938583092,"rawValue":1316.024938583092,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=724","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-980634167f4eab6455ee7ae7","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1296.1252619476961,"rawValue":1296.1252619476961,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=436","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-ac42a1cf8cde9b902ac7e2de","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1268.9507176368982,"rawValue":1268.9507176368982,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=32","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-8379f773a6e8daec48021d3c","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1279.0604106972467,"rawValue":1279.0604106972467,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=916","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-833e5a38c1303c644da6f854","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1262.4022278899602,"rawValue":1262.4022278899602,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=550","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-cd0a75a10c612354bf97da72","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1254.1527858213894,"rawValue":1254.1527858213894,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=362","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-3bdd82f3fb48ef042686c8ec","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1220.252084629971,"rawValue":1220.252084629971,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=859","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-d740c8af82094bf80561bea0","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1319.3826244267464,"rawValue":1319.3826244267464,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=419","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-ffcd303a2c9b747aec39093c","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1323.7480829167566,"rawValue":1323.7480829167566,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=720","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-6ce1c123df322be28494bf98","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1354.986672617962,"rawValue":1354.986672617962,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=195","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-b768c875b79b3bc3217dfb8b","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1287.1952109984525,"rawValue":1287.1952109984525,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=340","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-f61d7b8e4320a5422fd46c66","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1300.110262572361,"rawValue":1300.110262572361,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=904","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-426c848b5b4291c58e4a38f6","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1291.3320711674387,"rawValue":1291.3320711674387,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=20","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-ccf23ae6f789d15102681bdd","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1279.3052179019526,"rawValue":1279.3052179019526,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=825","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-d6e7687dbe8914215a1e8886","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1284.489086959617,"rawValue":1284.489086959617,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=536","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-fff757d3843cc5c3ad299e8c","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1343.6767036943584,"rawValue":1343.6767036943584,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=200","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-9cd0fe95a63351e4e52342ec","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1251.3036843705388,"rawValue":1251.3036843705388,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=364","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":12,"sourceMetricId":"elo"},{"id":"pub-32284efd1ef4355bbbea52eb","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1258.9378101102452,"rawValue":1258.9378101102452,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=554","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":13,"sourceMetricId":"elo"},{"id":"pub-d233dee4a74012014d3d467a","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1254.80414442906,"rawValue":1254.80414442906,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=43","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":14,"sourceMetricId":"elo"},{"id":"pub-54b0aae2b0e8bdf15aab6495","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1260.7183466538447,"rawValue":1260.7183466538447,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=927","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":15,"sourceMetricId":"elo"},{"id":"pub-7ba7bb93d01af7493c4352f9","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1257.627645091559,"rawValue":1257.627645091559,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=459","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":16,"sourceMetricId":"elo"},{"id":"pub-fe47361eb17a522c4c5dfb14","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1261.959039824833,"rawValue":1261.959039824833,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=758","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":17,"sourceMetricId":"elo"},{"id":"pub-0dcaead144c5a13a66d57627","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1284.0378835814013,"rawValue":1284.0378835814013,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=666","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-58f5aa336027e5b239b84233","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1253.19540869069,"rawValue":1253.19540869069,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=157","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"captioning","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-75e6dc1a9f77977795c943b6","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1284.3454563531582,"rawValue":1284.3454563531582,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=821","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-d6263e61bbc7b8a0b7ed805a","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1296.290374421956,"rawValue":1296.290374421956,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=734","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-2f7b224e85824580ff10c597","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1264.6581191073087,"rawValue":1264.6581191073087,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=354","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-550c155e739084ecf33fed2a","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1270.5106909568983,"rawValue":1270.5106909568983,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=544","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-82fd16dbfa09f17595b7b579","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1273.8052118537719,"rawValue":1273.8052118537719,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=918","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-7932013c42fb025f187c195b","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1266.7468257269916,"rawValue":1266.7468257269916,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=34","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-324c292c4dde3ce1f3b60a4b","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1317.8025399585476,"rawValue":1317.8025399585476,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=213","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-06e518aab32aff9be358df99","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1281.092411400433,"rawValue":1281.092411400433,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=444","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-0932b269597eb42b2e6b1098","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1348.6841917887145,"rawValue":1348.6841917887145,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=197","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-813c50361b22d0f9a66159db","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1284.8125957988436,"rawValue":1284.8125957988436,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=535","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-588b30a08b8582975e7c2cdd","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1310.1020101239699,"rawValue":1310.1020101239699,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=729","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-bdedf861b87fb800c1b3f056","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1278.8400373640645,"rawValue":1278.8400373640645,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=828","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-aee7b69882e18e2f45869682","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1297.7564823939383,"rawValue":1297.7564823939383,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=433","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-120fc66341c5ab6396249b53","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1290.8729564438972,"rawValue":1290.8729564438972,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=908","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-209efe70c3b8df34f810d221","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1280.7346998967805,"rawValue":1280.7346998967805,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=25","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-e49d7f616402fcfe66876354","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1265.2797344652085,"rawValue":1265.2797344652085,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=353","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-02be7129a863131848a92fdf","canonicalModelId":"xiaomi/mimo-v2-omni@current","modelName":"MiMo V2 Omni","modelRef":"mimo-v2-omni","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1281.2103866483121,"rawValue":1281.2103866483121,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=228","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8517a59806c74ab041364365","canonicalModelId":"xiaomi/mimo-v2-omni@current","modelName":"MiMo V2 Omni","modelRef":"mimo-v2-omni","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1255.1672936816294,"rawValue":1255.1672936816294,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=462","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-891295d0d8957f613d5c4518","canonicalModelId":"xiaomi/mimo-v2-omni@current","modelName":"MiMo V2 Omni","modelRef":"mimo-v2-omni","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1253.2921022521436,"rawValue":1253.2921022521436,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=762","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-a3d5cd67f462cf4b1fe84b43","canonicalModelId":"xiaomi/mimo-v2-omni@current","modelName":"MiMo V2 Omni","modelRef":"mimo-v2-omni","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1241.0158810190953,"rawValue":1241.0158810190953,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=569","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-b96bcbdf52cbc36e72820b83","canonicalModelId":"xiaomi/mimo-v2-omni@current","modelName":"MiMo V2 Omni","modelRef":"mimo-v2-omni","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1238.9233609436715,"rawValue":1238.9233609436715,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=944","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-685bbeecc958684e2f078521","canonicalModelId":"xiaomi/mimo-v2-omni@current","modelName":"MiMo V2 Omni","modelRef":"mimo-v2-omni","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1228.8261260156598,"rawValue":1228.8261260156598,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=62","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-297e95708898766f2f5a6854","canonicalModelId":"xiaomi/mimo-v2-omni@current","modelName":"MiMo V2 Omni","modelRef":"mimo-v2-omni","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1182.0435256255978,"rawValue":1182.0435256255978,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=871","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-2ba52b585da563ac672ab5da","canonicalModelId":"xiaomi/mimo-v2-omni@current","modelName":"MiMo V2 Omni","modelRef":"mimo-v2-omni","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1188.245002592475,"rawValue":1188.245002592475,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=395","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-fff67674db17da2d59350a90","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1328.613451384389,"rawValue":1328.613451384389,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=206","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-5251e4990e23f46a628829fd","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1195.7061799232793,"rawValue":1195.7061799232793,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=695","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-fec608d411ee66345f54ff21","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1271.8996609080855,"rawValue":1271.8996609080855,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=448","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-79348326b53c3dbe7aba9a55","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1262.135794150942,"rawValue":1262.135794150942,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=551","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-4adb8e94d772f8ec93c1c5f6","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1259.012296049082,"rawValue":1259.012296049082,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=928","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-de7e809df279a6bcc240b371","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1248.5602985970718,"rawValue":1248.5602985970718,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=48","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-81ade7c0248eed40878aba21","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1266.2051212157517,"rawValue":1266.2051212157517,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=755","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-52a99d0211665e00b2002b30","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1225.1121806112103,"rawValue":1225.1121806112103,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=853","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-1d6dc4e730f3d28d425f1b6d","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1223.9541695976154,"rawValue":1223.9541695976154,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=381","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-946318fec2eb4d824f373c09","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1319.7348565813618,"rawValue":1319.7348565813618,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=210","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c73d54049ba2dd92407d50a4","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1265.6258205852478,"rawValue":1265.6258205852478,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=547","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-1abc93758580fe6201f976d8","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1259.189231040419,"rawValue":1259.189231040419,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=840","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-55779c0360d973bf3b3b75ba","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1264.213995432525,"rawValue":1264.213995432525,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=925","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-a2cb7ae617d76ef673e429f1","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1255.1367541779075,"rawValue":1255.1367541779075,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=42","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-877f300dc22d4c25c6df53e7","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1265.633147559231,"rawValue":1265.633147559231,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=453","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-69575293b6af5e62619176a8","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1270.1936130461481,"rawValue":1270.1936130461481,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=753","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-648d14867e5ef8e022f10813","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1238.8376446035531,"rawValue":1238.8376446035531,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=374","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-473c0c24b969a35f14c3f4bf","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1300.6831570835932,"rawValue":1300.6831570835932,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=218","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-1767733a06933dc3ad96ae48","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1250.7123541477056,"rawValue":1250.7123541477056,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=843","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-1f81ffef01ca3c07d546a8ad","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1246.7276612449236,"rawValue":1246.7276612449236,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=563","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-b0130e595d789a1491e426e5","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1245.589759515466,"rawValue":1245.589759515466,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=52","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-140104f8a72830dfc591793e","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1251.5525820633393,"rawValue":1251.5525820633393,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=936","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-cbc54a52ee8724500d3c999f","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1260.2362448052645,"rawValue":1260.2362448052645,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=759","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-b503c3f4fccbbd4b9605179f","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1226.2761926535172,"rawValue":1226.2761926535172,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=379","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-45c6dd76bd4556d70033cc10","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1250.807278997042,"rawValue":1250.807278997042,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=465","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-a80e88deb9cb6bd5bee51605","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1196.1628693715452,"rawValue":1196.1628693715452,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=694","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-170d69c4ea012e1b5280978f","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1307.7550228241437,"rawValue":1307.7550228241437,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=215","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-a2dc0a80b1511baa9158e358","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1221.721067038316,"rawValue":1221.721067038316,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=856","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-cc8548cdd8c39cfcb6e8d490","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1254.8602796340763,"rawValue":1254.8602796340763,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=760","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-4dfc61902b108a6bd2b2153a","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1244.6977423948695,"rawValue":1244.6977423948695,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=566","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-e5c7ccb724f96f6cde17976c","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1241.322690881333,"rawValue":1241.322690881333,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=55","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-937a112201404671eb0d89d8","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1248.9791828179539,"rawValue":1248.9791828179539,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=939","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-298949554b8bcdefad45f193","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1250.3729374040126,"rawValue":1250.3729374040126,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=466","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-7884fecb70f97dc5c11016d4","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1223.56785471743,"rawValue":1223.56785471743,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=382","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-e8bff0432686733cdc8f2c08","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1277.5273176078551,"rawValue":1277.5273176078551,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=669","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"entity_recognition","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-934cb1ca366dc1e66b43dbb7","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1270.788610518163,"rawValue":1270.788610518163,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=348","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-e1f26a98017be70d4efaf4bb","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1324.3954494186592,"rawValue":1324.3954494186592,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=207","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-a76605ed7c95725b81d21b62","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1272.2559184358659,"rawValue":1272.2559184358659,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=542","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-3d5257ba0675b0d8b22d6439","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1288.1532616900909,"rawValue":1288.1532616900909,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=736","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-1a3f446f18ef7ccac3c4e0cb","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1270.3556461076255,"rawValue":1270.3556461076255,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=834","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-c40ab8047e5ac32ed704917d","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1273.1663913122293,"rawValue":1273.1663913122293,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=919","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-f50dff7287c226e2bfc2ee0a","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1264.8198799115057,"rawValue":1264.8198799115057,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=36","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-65d9a6a6989a5823895ca2af","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1280.88441786539,"rawValue":1280.88441786539,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=445","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-184e572eb87705ed13f8c713","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1287.8850369104794,"rawValue":1287.8850369104794,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=818","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-7bfe573395c60eda4ac4ac44","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1287.7995597979207,"rawValue":1287.7995597979207,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=910","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-8327ad83008d58ad242de7a0","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1280.597168003692,"rawValue":1280.597168003692,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=27","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-68f547aed5bbb50b329abfd3","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1268.346504123658,"rawValue":1268.346504123658,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=350","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-4afe8cc190b11c0a7069cab1","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1274.6474959039533,"rawValue":1274.6474959039533,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=539","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-6afc0b7da67994ac44855e8a","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1290.1653710018306,"rawValue":1290.1653710018306,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=439","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-5c03537c0fc538cf13c162ce","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1323.080457534263,"rawValue":1323.080457534263,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=209","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-ef3dbfa86f0daed725861200","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1287.0197766302144,"rawValue":1287.0197766302144,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=738","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-b0de2aff4268cc30a13474a8","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1280.7069192307151,"rawValue":1280.7069192307151,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=26","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-bb9e381d9defad90f510c003","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1295.6609945028163,"rawValue":1295.6609945028163,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=437","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-ba342c828e283e4d52b295d9","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1283.2063334985266,"rawValue":1283.2063334985266,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=914","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-d24fd33a2541795b9b9f30fc","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1269.0314203901398,"rawValue":1269.0314203901398,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=545","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-5ba48893b8a57c017cc64d22","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1271.7680370261946,"rawValue":1271.7680370261946,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=752","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-5e7cda915be0b7bb11eb65c2","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1247.2306170456689,"rawValue":1247.2306170456689,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=370","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-e3c87c421155e1c88aaea7e8","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1314.938580023343,"rawValue":1314.938580023343,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=4","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-12ece2d189d010ffafbb78f4","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1337.8449552750146,"rawValue":1337.8449552750146,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=710","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"homework","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-4b345e036b443f000b23df07","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1331.8903493484236,"rawValue":1331.8903493484236,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=414","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"diagram","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-3b0067c954cdcd03adfb2b3d","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1324.0644823924113,"rawValue":1324.0644823924113,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=889","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"ocr","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-b7a5a00699b4796532fe7135","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1310.458199904866,"rawValue":1310.458199904866,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=518","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"english","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-5af1e6d57e0ba8ee3ae1e6a6","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1314.9617396608874,"rawValue":1314.9617396608874,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=330","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"creative_writing_vision","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-ba48eec9a9563ea984753b5d","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1294.1556781586576,"rawValue":1294.1556781586576,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=814","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"humor","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-a1169fd86519f57742bf022d","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-vision","benchmarkName":"Arena Vision","benchmarkVersionHint":"arena-vision@latest","metricId":"arena_score_bt","value":1355.347785454357,"rawValue":1355.347785454357,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=194","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"vision","category":"chinese","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"model"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-a0077754bce50747b48f8f16","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1666.4803631887767,"rawValue":1666.4803631887767,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=281","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-28d118fc4b7a613f5e3889aa","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1623.1497118889843,"rawValue":1623.1497118889843,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=119","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-600c9682408cecc45a896182","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1626.2590219836861,"rawValue":1626.2590219836861,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=167","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-577eab6ceb0e07bd38e7c9e5","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1629.661824704901,"rawValue":1629.661824704901,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=402","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-14c8cea38b26a7e5cda2f51e","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1626.2590219836861,"rawValue":1626.2590219836861,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=5","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-ab11c844fd4721ce2eb18fdd","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1324.1907410858214,"rawValue":1324.1907410858214,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=483","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-94d49cd628fc8feb80db7f9f","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1327.4094602355049,"rawValue":1327.4094602355049,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=93","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-5f0ab35fa131fb97f8e329d2","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1327.4094602355049,"rawValue":1327.4094602355049,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=255","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-5263c8c20a0d0f68ba1cf340","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1315.340717059536,"rawValue":1315.340717059536,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=374","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-58ae30cd39bde23605404ab8","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1481.8384219267882,"rawValue":1481.8384219267882,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=320","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-e4fa4a063a4cecad95f16040","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1468.0777023872317,"rawValue":1468.0777023872317,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=203","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-cc2b38dc109e18c6a0abe327","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1456.3995394700153,"rawValue":1456.3995394700153,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=438","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-a4f22b81df54bf4ba66e0f79","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1468.0777023872317,"rawValue":1468.0777023872317,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=41","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-b851d201941cab101f62c206","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-high-32k","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1511.5803414416328,"rawValue":1511.5803414416328,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=313","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-aa8acf84715874de7476ed22","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-high-32k","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1494.3932987922449,"rawValue":1494.3932987922449,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=34","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-1c7139c8285cb3036d1cef57","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-high-32k","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1494.3932987922449,"rawValue":1494.3932987922449,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=196","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-5cbe7ca4162f9066f0611be6","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-high-32k","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1480.0731797318567,"rawValue":1480.0731797318567,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=433","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-587d9786185483ee03d1bbff","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1539.349142690242,"rawValue":1539.349142690242,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=130","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-90cead1b76a6cae8ec24c9f6","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1550.0478852675992,"rawValue":1550.0478852675992,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=296","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-912e69143a113afa7e500601","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1536.273082073974,"rawValue":1536.273082073974,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=184","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-177a9e2d3bb1bcd10ebe2c0b","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1536.273082073974,"rawValue":1536.273082073974,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=22","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-960bb971202ae5582b215be7","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1528.6853512923549,"rawValue":1528.6853512923549,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=420","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-97f979395a32687b8b778e53","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1541.2452138974932,"rawValue":1541.2452138974932,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=129","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-fa181f8953d55b9c39815825","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1559.2067277690508,"rawValue":1559.2067277690508,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=293","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-aa64bfe98c47dc4f37046f8d","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1545.6803042848985,"rawValue":1545.6803042848985,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=179","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-52afa7a513e9071161e9a293","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1545.6803042848985,"rawValue":1545.6803042848985,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=17","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-24138e48c33a9148637f5175","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1539.2014452376384,"rawValue":1539.2014452376384,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=416","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-5e64b757f820197935ea1c6f","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1564.300417312133,"rawValue":1564.300417312133,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=126","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-77985bc120238c8c7809d8f5","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1559.4957732437688,"rawValue":1559.4957732437688,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=292","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-b26239b3f17451063c7b5469","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1557.5168669531042,"rawValue":1557.5168669531042,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=14","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-60fffaf7a631b0fc50e1cf1c","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1557.5168669531042,"rawValue":1557.5168669531042,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=176","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-150a44ac5b383123443eccd9","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1554.6020279194017,"rawValue":1554.6020279194017,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=413","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-5635c3b901200f66c832fb63","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1575.8252930757917,"rawValue":1575.8252930757917,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=122","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-54a73eef52d403897d2fe5c0","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1555.0847118649995,"rawValue":1555.0847118649995,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=294","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-d45661d9c661248c82c0fe16","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1557.7123580122975,"rawValue":1557.7123580122975,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=412","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-374a996952e57506cf7fa831","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1556.695153981578,"rawValue":1556.695153981578,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=177","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-023c2872d86f9fb16d6a934a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1556.695153981578,"rawValue":1556.695153981578,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=15","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-eb7137e9b3ed0fb4e0e3f3cb","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1541.3594837146388,"rawValue":1541.3594837146388,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=415","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-7c8e3dff311f15d4352e9081","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1539.1006570198726,"rawValue":1539.1006570198726,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=180","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-4902314580f9f5c84dc173d5","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1539.1006570198726,"rawValue":1539.1006570198726,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=18","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-b4ee53a2cc752a9d17d8ede3","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1527.232355179835,"rawValue":1527.232355179835,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=307","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-b2d7e993f8a3882270ffbaa5","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1562.9855618932615,"rawValue":1562.9855618932615,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=410","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-95d113d47255e89b36ca5c6f","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1563.4925116674042,"rawValue":1563.4925116674042,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=175","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-58ca249b8e5d8a3c1df4118f","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1563.4925116674042,"rawValue":1563.4925116674042,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=13","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-3ebe46ae06ab0a034733b772","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1554.484277741286,"rawValue":1554.484277741286,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=295","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-40003d809af84f318424bf26","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1680.8083757859047,"rawValue":1680.8083757859047,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=399","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-893e236fc0d016763659de31","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1662.5934919996598,"rawValue":1662.5934919996598,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=3","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-3d744cca97c9e64fe28a293e","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1648.1123955670514,"rawValue":1648.1123955670514,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=283","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-3cff507069f9de4457484217","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1662.5934919996598,"rawValue":1662.5934919996598,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=165","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-f24237a90d7cf973a7100983","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1685.0219335164438,"rawValue":1685.0219335164438,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=280","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-ec0e93df7f6c17f2d0ea1747","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1709.1575588894566,"rawValue":1709.1575588894566,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=397","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-bef00c97fc9cff9a55725ddc","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1690.6395713997163,"rawValue":1690.6395713997163,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=162","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-7b7365af989972b228c0ce78","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1690.6395713997163,"rawValue":1690.6395713997163,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=0","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-13243749e6268935a9ad567d","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1664.3882751396995,"rawValue":1664.3882751396995,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=118","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-6a682c7770fdc50911c18f94","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1383.8541311856036,"rawValue":1383.8541311856036,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=464","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-28f6f01740734984f3d1ec13","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1385.5090474921917,"rawValue":1385.5090474921917,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=73","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-1213c9ffedef880964eb0166","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1385.5090474921917,"rawValue":1385.5090474921917,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=235","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-77146f8e9f1de134e2e4d82c","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1390.2739047408186,"rawValue":1390.2739047408186,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=354","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-69d185ca5a2ef26097308d2a","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929-high-32k","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1385.519914415382,"rawValue":1385.519914415382,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=463","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-7bbf8d0e78009fd9df959e65","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929-high-32k","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1392.3609892159568,"rawValue":1392.3609892159568,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=68","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-254d6ba36156ba935d137875","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929-high-32k","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1392.3609892159568,"rawValue":1392.3609892159568,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=230","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-f0426ac7a01edf476184248e","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929-high-32k","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1398.7746311206224,"rawValue":1398.7746311206224,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=349","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-4d5049a788b9ba2a9ed9ba6a","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1529.1743836735386,"rawValue":1529.1743836735386,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=134","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-dffb3c16b8b5a3da859c56d0","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1522.467769386207,"rawValue":1522.467769386207,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=186","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-8f7093e28e46cdd10aef5146","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1522.467769386207,"rawValue":1522.467769386207,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=24","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-dfdf760e35ef66c0ddd9ca6f","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1529.0990045958115,"rawValue":1529.0990045958115,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=306","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-5362d11832dde929b5540b04","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1511.8758873505844,"rawValue":1511.8758873505844,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=424","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-792a7dc521282a838d4c3bfa","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1542.606007160266,"rawValue":1542.606007160266,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=128","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-1fb85aa22dfaed4c16b523d4","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1542.5206202560817,"rawValue":1542.5206202560817,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=414","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-af197701a3096a0769a28011","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1538.5842623430603,"rawValue":1538.5842623430603,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=21","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-083d4e9db9d3550d543301d5","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1538.5842623430603,"rawValue":1538.5842623430603,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=183","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-f17813b4e08c037a68cd512a","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1522.3186008214707,"rawValue":1522.3186008214707,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=309","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-90d7ef73ec5a9cc015608e57","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1335.3422141277977,"rawValue":1335.3422141277977,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=481","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-0ba1d0f00081e32e4e293cb8","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1324.5403629748857,"rawValue":1324.5403629748857,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=256","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-0add8cef7053fa1ed22d8263","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1324.5403629748857,"rawValue":1324.5403629748857,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=94","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-6558aec4066dc1f58fc042dc","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1303.936052735041,"rawValue":1303.936052735041,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=377","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-7188981c01735b13638c149f","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"deepseek-v3.2-exp","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1271.7662103406499,"rawValue":1271.7662103406499,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=100","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-415725d521532d5f54a957a3","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"deepseek-v3.2-exp","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1271.7662103406499,"rawValue":1271.7662103406499,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=262","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-25b9afc3f05b8c07ff66546a","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"deepseek-v3.2-exp","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1281.774112303272,"rawValue":1281.774112303272,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=381","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-580610a6a21a693ec3f6fc58","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2-thinking","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1349.8597570092388,"rawValue":1349.8597570092388,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=476","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-1a093866b8689bca5df17e47","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2-thinking","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1368.5876510360758,"rawValue":1368.5876510360758,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=360","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-66318a921cf66c773a7d40f1","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2-thinking","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1360.3128404975844,"rawValue":1360.3128404975844,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=243","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-36dd4787028b44ad36877292","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-v3.2-thinking","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1360.3128404975844,"rawValue":1360.3128404975844,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=81","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-18e6328029aa1800670cc662","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1583.5629500923421,"rawValue":1583.5629500923421,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=408","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-5f662fcc22159f7058520800","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1579.0993765147234,"rawValue":1579.0993765147234,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=174","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-2f319475fe37c7243da64399","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1579.0993765147234,"rawValue":1579.0993765147234,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=12","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-54a7265602a4ad92799e0dd0","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1542.27813546253,"rawValue":1542.27813546253,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=298","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-e791bed498375b34c913012c","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-high-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1422.5202197873225,"rawValue":1422.5202197873225,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=453","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-cafef44bd84b3006c259219b","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-high-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1431.069014492644,"rawValue":1431.069014492644,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=57","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-9f187821ef3a99a08d5e1bc3","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-high-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1431.069014492644,"rawValue":1431.069014492644,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=219","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-09696af3a40b9e12fd48aa0b","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-high-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1422.9704052212535,"rawValue":1422.9704052212535,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=340","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-d860309764ed1e66975eed7d","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1439.2137888468844,"rawValue":1439.2137888468844,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=443","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-6e2415f8885566438278d486","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1445.4554435536177,"rawValue":1445.4554435536177,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=210","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-5e148ba8679b650bfe451bd7","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1445.4554435536177,"rawValue":1445.4554435536177,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=48","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-4d791da771c410de0043dfdb","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1436.7593568233792,"rawValue":1436.7593568233792,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=333","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-901c6d7a4864baa30be23b90","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-high-20260813","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1588.6753688261556,"rawValue":1588.6753688261556,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=407","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-d132cf5488672826af4c680b","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-high-20260813","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1563.003289949429,"rawValue":1563.003289949429,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=291","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-9756fa882cdf906b4fc4fb83","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-high-20260813","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1581.6418400239618,"rawValue":1581.6418400239618,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=173","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-2bf3c4625ef2ad25b6d0125c","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-high-20260813","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1581.6418400239618,"rawValue":1581.6418400239618,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=11","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-f725d92533b7fc2b04ecaf8d","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro-high-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1463.8042124059687,"rawValue":1463.8042124059687,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=204","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-759c221f7ccbe17a9f116a56","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro-high-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1463.8042124059687,"rawValue":1463.8042124059687,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=42","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-4069a7243578690f69f6fcce","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro-high-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1454.6125978774453,"rawValue":1454.6125978774453,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=439","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-4dcd94fc55c902d7eef8ba26","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro-high-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1471.740281364178,"rawValue":1471.740281364178,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=323","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-68258c164a897ee3a74e6ddb","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1283.6824608232687,"rawValue":1283.6824608232687,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=161","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-9b5a7fe96d3a2e447c3b9112","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1234.8747938414745,"rawValue":1234.8747938414745,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=389","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-ba3f2ab2a1a251a9d8a46c28","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1225.6866415067607,"rawValue":1225.6866415067607,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=272","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-6d7d476ee582ccb3340e13fc","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1225.6866415067607,"rawValue":1225.6866415067607,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=110","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-f9b38676175a0b0196e2e56b","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1480.5198401926805,"rawValue":1480.5198401926805,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=143","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-0fb57f64a5a3e9e8a14ea929","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1493.2614015377562,"rawValue":1493.2614015377562,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=317","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-6a70888551f7db985ef3e781","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1436.495850185246,"rawValue":1436.495850185246,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=444","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-6292b59e643df4352d8298f4","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1446.4556668949795,"rawValue":1446.4556668949795,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=47","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-1ac05ddb51726e56cf7537eb","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1446.4556668949795,"rawValue":1446.4556668949795,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=209","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-4b25372ff623d034fc5924bd","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1493.6890751917033,"rawValue":1493.6890751917033,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=142","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-d1df4677fb6a7ba6494d7505","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1519.0834325869387,"rawValue":1519.0834325869387,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=312","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-d734714bba1d517c289228ed","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1498.5856161377683,"rawValue":1498.5856161377683,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=195","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-b03bad554b878e035d88ac90","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1498.5856161377683,"rawValue":1498.5856161377683,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=33","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-6d4148fa89a6874a3298cfe2","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1491.539311545379,"rawValue":1491.539311545379,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=430","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-2699ada5a6f6059a04a8f1b2","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1443.7875790280625,"rawValue":1443.7875790280625,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=149","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-9c319f416b1ceb61c2b75edc","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1449.1411432227515,"rawValue":1449.1411432227515,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=46","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-7e50452a9446a6e3e62f3e9f","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1449.1411432227515,"rawValue":1449.1411432227515,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=208","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-58d0de43e1f396e8647eca55","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1425.1457138983424,"rawValue":1425.1457138983424,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=451","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-c7c84fc177960e7bce2178cb","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1499.011477726488,"rawValue":1499.011477726488,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=140","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-47d22e7ff3cc325a44129991","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1523.8905440947433,"rawValue":1523.8905440947433,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=308","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-dc8479bbdaab4dd23c3566d7","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1489.741632583298,"rawValue":1489.741632583298,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=197","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-5dff0d1ff95ac289fa7d9aed","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1489.741632583298,"rawValue":1489.741632583298,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=35","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-582a942f3438bb169ccc52eb","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-medium","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1482.952749144996,"rawValue":1482.952749144996,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=432","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-91438fbdddd3fc145182ac2a","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1544.437428907048,"rawValue":1544.437428907048,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=127","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-ea6fb39cc8d1edfec6108c43","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1538.9071087861485,"rawValue":1538.9071087861485,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=20","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-af422b3a2ee242cd5598ec34","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1538.9071087861485,"rawValue":1538.9071087861485,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=182","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-42328637bee9bcf2afb9b6d8","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1529.8411576375765,"rawValue":1529.8411576375765,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=418","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-15f16344d960f590f18348c4","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1535.4738557700312,"rawValue":1535.4738557700312,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=304","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-cfd8cd9df2232ee76509b63d","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1622.34702959783,"rawValue":1622.34702959783,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=286","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8d41218517459103f97d11fe","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1587.044813853989,"rawValue":1587.044813853989,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=9","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-6cc35a212addd733f84799d2","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1587.044813853989,"rawValue":1587.044813853989,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=171","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-a1b0af00e339346a8a409c76","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1581.4200321442174,"rawValue":1581.4200321442174,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=409","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-88f7329fc657ddc33d907db3","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1446.2888781758202,"rawValue":1446.2888781758202,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=330","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-09f1b20408c5c255a43663cf","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1428.757810074879,"rawValue":1428.757810074879,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=448","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-5b3d88c93ccad8293428b7bf","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1434.289758171793,"rawValue":1434.289758171793,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=55","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-517b9633f55c60b599ef8722","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1434.289758171793,"rawValue":1434.289758171793,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=217","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-0158143df2becd022771cb97","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1454.1117867061448,"rawValue":1454.1117867061448,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=329","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-9f4495f7ceb70c192e4ee0c5","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1435.4584267142257,"rawValue":1435.4584267142257,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=216","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-01d1a0b6426a9318afd6e4de","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1435.4584267142257,"rawValue":1435.4584267142257,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=54","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-4edf74e45d86fecd18714dac","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1424.6289525106836,"rawValue":1424.6289525106836,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=452","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-a82c78acead6d334c6fdba92","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1508.9259030653723,"rawValue":1508.9259030653723,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=30","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-634eb959a04e8bb89b7e1828","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1504.6689943291394,"rawValue":1504.6689943291394,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=427","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-585f5b708fba1e957522bffb","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1508.9259030653723,"rawValue":1508.9259030653723,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=192","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-bbffb41a3bda9f9591dcb679","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1519.7441056250213,"rawValue":1519.7441056250213,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=311","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-c61e5792588c2e4ce409957b","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1593.532925524643,"rawValue":1593.532925524643,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=406","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-4df50c225e9e5cb8fba5d431","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1582.0912797071308,"rawValue":1582.0912797071308,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=10","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-385947df459796b0dfbff214","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1582.0912797071308,"rawValue":1582.0912797071308,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=172","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-588f2dab6a6a9937fc25e229","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1536.5926280216395,"rawValue":1536.5926280216395,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=301","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-a56ec55da46855531df8d1fa","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1598.6001848847525,"rawValue":1598.6001848847525,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=7","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8dd04cb44bd39ecefe09c5f6","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1609.2520822615816,"rawValue":1609.2520822615816,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=404","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-0135e6180c96d8f8ff4ac530","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1598.6001848847525,"rawValue":1598.6001848847525,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=169","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-da29e3b24f31ecbaabf2a117","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1577.091365387705,"rawValue":1577.091365387705,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=289","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-4ebd14b38ef701842c65229f","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1423.425383722039,"rawValue":1423.425383722039,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=153","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-dc3b6ecb20f7edc170895922","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1399.8264781078112,"rawValue":1399.8264781078112,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=226","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-23f775f249091f24b1a8a9c5","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1399.8264781078112,"rawValue":1399.8264781078112,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=64","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-e37a460a58467a7edc782a56","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1383.83048846566,"rawValue":1383.83048846566,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=465","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-94b0b56be1c6149172902ce7","canonicalModelId":"zai/glm-5v-turbo@2026-04-01","modelName":"GLM-5V-Turbo","modelRef":"glm-5v-turbo","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1364.6644015609772,"rawValue":1364.6644015609772,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=362","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-e945f9d4bf903ea8ee5a2eb8","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-medium","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1428.9187308641385,"rawValue":1428.9187308641385,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=335","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-fd2a24dbdd47639de0c2630a","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-medium","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1419.1398643855996,"rawValue":1419.1398643855996,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=220","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-e29d675a1a84941719db9aa6","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-medium","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1419.1398643855996,"rawValue":1419.1398643855996,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=58","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-a447a837ce9a5ae68227adb1","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1344.786912963157,"rawValue":1344.786912963157,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=159","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-a7baf1854bb47aaa76e8976e","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1364.1835378758472,"rawValue":1364.1835378758472,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=363","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-c2d9209478d5e960aebbc7f1","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1340.632179382849,"rawValue":1340.632179382849,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=250","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-9e828425f309265621f21d92","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1340.632179382849,"rawValue":1340.632179382849,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=88","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-aeae0929a2c3b98903a556ae","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1304.5400830339133,"rawValue":1304.5400830339133,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=486","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-d71dcd0d864d01e84e247583","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"gpt-5.1-codex","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1345.8733070042335,"rawValue":1345.8733070042335,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=370","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-bea2af75747d983eabdb0adf","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"gpt-5.1-codex","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1336.2778561874031,"rawValue":1336.2778561874031,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=253","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-b04d885cf2ae5c3ce0ece77e","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"gpt-5.1-codex","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1336.2778561874031,"rawValue":1336.2778561874031,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=91","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-58a5f67cc025d2afb3b708e6","canonicalModelId":"openai/gpt-5.1-codex-mini@2025-11-17","modelName":"GPT-5.1 Codex Mini","modelRef":"gpt-5.1-codex-mini","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1253.6184669539425,"rawValue":1253.6184669539425,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=385","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-318ce3bc3af113b875505a49","canonicalModelId":"openai/gpt-5.1-codex-mini@2025-11-17","modelName":"GPT-5.1 Codex Mini","modelRef":"gpt-5.1-codex-mini","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1243.7391069949672,"rawValue":1243.7391069949672,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=267","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-070ad373581da8f294004efb","canonicalModelId":"openai/gpt-5.1-codex-mini@2025-11-17","modelName":"GPT-5.1 Codex Mini","modelRef":"gpt-5.1-codex-mini","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1243.7391069949672,"rawValue":1243.7391069949672,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=105","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-b4bde459b2d610572af56674","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1419.758619176206,"rawValue":1419.758619176206,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=154","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-fca811f54e2c3ed9468f8413","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-medium","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1401.0621185609332,"rawValue":1401.0621185609332,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=347","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-919b22e39cedf40b080f8604","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-medium","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1391.014390336067,"rawValue":1391.014390336067,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=231","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-3ea0740e643f3b8782441220","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-medium","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1391.014390336067,"rawValue":1391.014390336067,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=69","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-370c04f673537a9fa15c7231","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1428.395452781471,"rawValue":1428.395452781471,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=336","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-b6264b7f3d3887dbfebaa362","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1417.810641177648,"rawValue":1417.810641177648,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=221","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-23f08737fb6656b29d6d09c0","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1417.810641177648,"rawValue":1417.810641177648,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=59","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-9c31d3be721ab51e6aaf1361","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1329.643790401107,"rawValue":1329.643790401107,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=482","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-dfd720a95b5233bf466b00ac","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1348.245079097252,"rawValue":1348.245079097252,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=369","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-e948f5f2699187e5a76a83b5","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1338.2841667454431,"rawValue":1338.2841667454431,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=90","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-08721cf40908c35093d1eaf9","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1338.2841667454431,"rawValue":1338.2841667454431,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=252","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-70543969743a3d4f5e96468f","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1444.2663125087988,"rawValue":1444.2663125087988,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=148","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c744b5db5d4b49c44e8b8ad5","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1427.3925039892774,"rawValue":1427.3925039892774,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=338","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-0aa3f555043ef375fe1ffb4c","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1397.9793187551898,"rawValue":1397.9793187551898,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=455","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-43d76f43745ea44f0d28750b","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1408.3779995790048,"rawValue":1408.3779995790048,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=60","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-3dd017d866ecf478ff6958af","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1408.3779995790048,"rawValue":1408.3779995790048,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=222","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-b35e8499a2354becb5dccbd8","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1419.8166295772658,"rawValue":1419.8166295772658,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=342","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-01da5ef65e22d8df28a9d1e2","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1354.5204484579363,"rawValue":1354.5204484579363,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=472","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-aa944e8f0f8978b69e188c31","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1368.9959519462375,"rawValue":1368.9959519462375,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=240","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-691f65be399bbc60bccc7fd9","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1368.9959519462375,"rawValue":1368.9959519462375,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=78","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-dfe2f597e7e151bffc42bb40","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1442.2774568784275,"rawValue":1442.2774568784275,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=150","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8de2a382ecaaa115f7074d16","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1423.2094096779504,"rawValue":1423.2094096779504,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=339","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-6f73130f09be67dec1cc7447","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1388.0096165628515,"rawValue":1388.0096165628515,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=459","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-c50406e4e7590958a88e2abd","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1390.556992986316,"rawValue":1390.556992986316,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=232","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":12,"sourceMetricId":"elo"},{"id":"pub-b8cca007d58df9397968e84f","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1390.556992986316,"rawValue":1390.556992986316,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=70","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":13,"sourceMetricId":"elo"},{"id":"pub-d6e0ee0bb9196c45004bafde","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1476.7572112847733,"rawValue":1476.7572112847733,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=321","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-807795cd40c2e0d071e03057","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1462.8963395174703,"rawValue":1462.8963395174703,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=205","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-0ed268695e6e009974f5b9f0","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1462.8963395174703,"rawValue":1462.8963395174703,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=43","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-4023cc7643c79283cbc32e3a","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-high (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1453.1340210108017,"rawValue":1453.1340210108017,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=441","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-6bfcb5a64d0adac3ca4c5ef9","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-medium (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1474.048924083781,"rawValue":1474.048924083781,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=322","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-a52c8de424ab3b34711a7d73","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-medium (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1429.1866596212149,"rawValue":1429.1866596212149,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=445","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-896ee6e64ca816533abe73b7","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-medium (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1442.1566799903471,"rawValue":1442.1566799903471,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=49","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-49f6eb884915dbabd2d7d6cb","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-medium (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1442.1566799903471,"rawValue":1442.1566799903471,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=211","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-620403d76fe410047ba04822","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1422.8173553543616,"rawValue":1422.8173553543616,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=341","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-5d968d78c74a760d88ee1dc7","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1387.0361208776533,"rawValue":1387.0361208776533,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=462","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-9a49a5e8f65753fbf06257fe","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1397.0815445066285,"rawValue":1397.0815445066285,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=229","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-52aae1e04e7df00ffb027ea2","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1397.0815445066285,"rawValue":1397.0815445066285,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=67","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-f66bcb3b3686fda44f6314c7","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5 (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1500.599108273734,"rawValue":1500.599108273734,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=139","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-cdc2f48280c8dcbca4e90d0c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5 (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1501.1631416341384,"rawValue":1501.1631416341384,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=314","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-fb2bc522311bb323a8fc996f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5 (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1457.5140563377497,"rawValue":1457.5140563377497,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=207","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":13,"sourceMetricId":"elo"},{"id":"pub-da0b9700dd8b122b7608eaa9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5 (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1447.234029142822,"rawValue":1447.234029142822,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=442","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":14,"sourceMetricId":"elo"},{"id":"pub-85e10562a9dafd9f278b5211","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5 (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1457.5140563377497,"rawValue":1457.5140563377497,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=45","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":15,"sourceMetricId":"elo"},{"id":"pub-983b734f32e60970553b4710","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1516.399261218979,"rawValue":1516.399261218979,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=136","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-7f3a5754a6859bda3cff786d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1539.9274492312652,"rawValue":1539.9274492312652,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=299","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-700c9b6de962d6b94ca2b42a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1486.1697842138751,"rawValue":1486.1697842138751,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=37","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-670e582fc746333379ca5b30","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1476.9618265794325,"rawValue":1476.9618265794325,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=434","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":11,"sourceMetricId":"elo"},{"id":"pub-2d008bb685e31b333fcea4b3","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-high (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1486.1697842138751,"rawValue":1486.1697842138751,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=199","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":12,"sourceMetricId":"elo"},{"id":"pub-653e81269db81930b92a16da","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1529.7172814531418,"rawValue":1529.7172814531418,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=133","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-69d78251118af24ba7924f68","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1544.6596061546925,"rawValue":1544.6596061546925,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=297","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-346aebad61cad329dc8ad9dd","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1500.3687592412523,"rawValue":1500.3687592412523,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=429","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-3172a18edae7f50e863721b0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1508.3502120258615,"rawValue":1508.3502120258615,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=32","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-14a6e7eaf59d01fb4166112f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1508.3502120258615,"rawValue":1508.3502120258615,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=194","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-c872900b2bc23f73678d2066","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1495.8214394408,"rawValue":1495.8214394408,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=141","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-b26b0de0090e68c2d35c3512","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1532.8442516988694,"rawValue":1532.8442516988694,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=305","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-90fcce7acbfc8fd771857b06","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1517.597333795525,"rawValue":1517.597333795525,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=190","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-8a3246e5f093f587bfedb30c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1517.597333795525,"rawValue":1517.597333795525,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=28","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-2e5f54e52e5ec86ee115db71","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1505.9475142861716,"rawValue":1505.9475142861716,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=426","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-403eeff24ad4c71328d0181f","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1606.4791387238115,"rawValue":1606.4791387238115,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=121","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-44a2ea67a334bb2ded113eed","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1624.9415689051705,"rawValue":1624.9415689051705,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=403","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-434d8c41c52dc4e70b6aa4e0","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1618.7118275567602,"rawValue":1618.7118275567602,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=6","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-42e7d3050c134f035cc4714f","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1618.7118275567602,"rawValue":1618.7118275567602,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=168","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-3caca9e58b0cab5bd1abbec1","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1618.2748501551378,"rawValue":1618.2748501551378,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=287","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-d1d392e0c4fccf49f435e55e","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1534.2537401820946,"rawValue":1534.2537401820946,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=132","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-8c7fb33abe8f0fdf65524cc3","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1536.0699760981693,"rawValue":1536.0699760981693,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=302","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-3c344f327fc8b742fd636563","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1519.8677186640261,"rawValue":1519.8677186640261,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=188","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-1e5f205634dce202ea758a17","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1519.8677186640261,"rawValue":1519.8677186640261,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=26","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-89bf76c2b3248e5afc3401f9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-xhigh (codex-harness)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1511.7410468620947,"rawValue":1511.7410468620947,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=425","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-b58d15d3d9736b63a681e633","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast-reasoning","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1170.2161843158192,"rawValue":1170.2161843158192,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=395","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-ce55eaa595b34dae3fbbc63b","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast-reasoning","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1161.8732122243282,"rawValue":1161.8732122243282,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=116","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-63692b37dab1e9fea6108c8c","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast-reasoning","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1161.8732122243282,"rawValue":1161.8732122243282,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=278","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-a7b5b82860d1faa6232c4ae0","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1574.3408563901694,"rawValue":1574.3408563901694,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=123","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-f44471f22c4fbc8003f3d5f2","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1576.772418486707,"rawValue":1576.772418486707,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=290","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-520a5cdafed9d0025772ae38","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1558.6065361696376,"rawValue":1558.6065361696376,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=411","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-ec47e9947e77193f8b51f77d","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1556.3617214249728,"rawValue":1556.3617214249728,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=16","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-6de4e8abe5f0896579f9f949","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1556.3617214249728,"rawValue":1556.3617214249728,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=178","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-d6e9d715972c09846d03dba1","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1638.3158059802884,"rawValue":1638.3158059802884,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=401","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-497d9f7fb0c216d733076d12","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1628.8418761621558,"rawValue":1628.8418761621558,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=4","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-430c25303d80cee6d65ba469","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1628.8418761621558,"rawValue":1628.8418761621558,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=166","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-314baa0e2dc0300b2ee08546","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6-high","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1631.454941066038,"rawValue":1631.454941066038,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=284","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-24552bb92a081c8ba30b3f2d","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1419.3949380571096,"rawValue":1419.3949380571096,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=155","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-3a08bb4e9573d63960565635","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1391.336100037613,"rawValue":1391.336100037613,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=458","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-74a65e715a1b3dc0a1696824","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1404.864925632702,"rawValue":1404.864925632702,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=224","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-69939919395603006e10798c","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1404.864925632702,"rawValue":1404.864925632702,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=62","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9,"sourceMetricId":"elo"},{"id":"pub-7740fbf64ddd93a80f0ae0fd","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-instant","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1414.20801751094,"rawValue":1414.20801751094,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=344","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10,"sourceMetricId":"elo"},{"id":"pub-f8c637d3ea6af02416523a67","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1439.8067470314134,"rawValue":1439.8067470314134,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=151","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-2ff1dcf6f17b22cb237591f8","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1428.7607397003983,"rawValue":1428.7607397003983,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=447","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-b17b5bb7638da6408e814066","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1435.9187347947436,"rawValue":1435.9187347947436,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=53","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-61302701644af69d52161aad","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1435.9187347947436,"rawValue":1435.9187347947436,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=215","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-b8acf944f1941635bdbd48cd","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5-thinking","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1429.8645129434199,"rawValue":1429.8645129434199,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=334","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-5cb6e47a257e9989f912e6cf","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1520.953328879154,"rawValue":1520.953328879154,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=135","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-6665f09160c9b9bd91630dac","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1508.7146104959886,"rawValue":1508.7146104959886,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=31","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-4bb5130133e4f27c4d24f30e","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1508.7146104959886,"rawValue":1508.7146104959886,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=193","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-09e7ec3920c8bee8d0593e89","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1503.6152719738125,"rawValue":1503.6152719738125,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=428","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-d79414915f9dd5731f4b7b02","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1494.9044004107018,"rawValue":1494.9044004107018,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=316","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-aee4bec6021ab9e9ac836256","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1510.6217076220032,"rawValue":1510.6217076220032,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=137","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-30bfd548553b40d3ef082ef0","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1472.3568398235943,"rawValue":1472.3568398235943,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=435","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-326be36383b910e5b6a531d9","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1472.646928867663,"rawValue":1472.646928867663,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=40","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-17bba06f3b9754f3d76de1b5","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1472.646928867663,"rawValue":1472.646928867663,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=202","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-aad5a1f10ce3d93e9163426e","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1461.9566376410835,"rawValue":1461.9566376410835,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=328","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-f1e4ee64f40b1aab7db81f6c","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1673.6961333978227,"rawValue":1673.6961333978227,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=1","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-dbc37933b276967436c01ab6","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1673.6961333978227,"rawValue":1673.6961333978227,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=163","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-66be88ae65b1610eb702f484","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1691.633912435193,"rawValue":1691.633912435193,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=398","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-00cfae1220900d9eeb130589","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1649.5392094087865,"rawValue":1649.5392094087865,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=282","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-a5acf693c5c1ed7bc780cad2","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1573.2857788791516,"rawValue":1573.2857788791516,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=125","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["heuristic_model_mapping","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-9e0a99fc876bb121763922a7","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (non-thinking)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1353.3448399219214,"rawValue":1353.3448399219214,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=367","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-c919707b3e2c6876e68fd8a5","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (non-thinking)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1309.3574712037303,"rawValue":1309.3574712037303,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=485","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-bfe30c3935b7eb3b4209cda1","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (non-thinking)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1330.1697248949574,"rawValue":1330.1697248949574,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=92","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-7d8da6bb50aad9ea3677fd6b","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (non-thinking)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1330.1697248949574,"rawValue":1330.1697248949574,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=254","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-d1bf61fb5dfa39a541e8a0c0","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (thinking)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1345.638787974788,"rawValue":1345.638787974788,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=371","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-da66c4e9b35f4e4a792e5d88","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (thinking)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1231.0262361783587,"rawValue":1231.0262361783587,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=493","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"sourceMetricId":"elo"},{"id":"pub-9556ebde6e61fba8112fd954","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (thinking)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1292.4940956740304,"rawValue":1292.4940956740304,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=260","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7,"sourceMetricId":"elo"},{"id":"pub-1fc261566dd053953c7c1de4","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash (thinking)","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1292.4940956740304,"rawValue":1292.4940956740304,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=98","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"sourceMetricId":"elo"},{"id":"pub-b5e30a7ce03aaffaf50cfb8e","canonicalModelId":"xiaomi/mimo-v2-pro@current","modelName":"MiMo V2 Pro","modelRef":"mimo-v2-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1439.7163716626642,"rawValue":1439.7163716626642,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=332","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-572be7bb31b4dc9b3b018fa0","canonicalModelId":"xiaomi/mimo-v2-pro@current","modelName":"MiMo V2 Pro","modelRef":"mimo-v2-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1425.7960537879328,"rawValue":1425.7960537879328,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=449","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-8e466987d2d9595ea59ad7d2","canonicalModelId":"xiaomi/mimo-v2-pro@current","modelName":"MiMo V2 Pro","modelRef":"mimo-v2-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1433.5094981427192,"rawValue":1433.5094981427192,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=56","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-50b434f18b8432d710694d8b","canonicalModelId":"xiaomi/mimo-v2-pro@current","modelName":"MiMo V2 Pro","modelRef":"mimo-v2-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1433.5094981427192,"rawValue":1433.5094981427192,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=218","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-6522d79c6cc527b968357fbc","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1437.7972257239153,"rawValue":1437.7972257239153,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=214","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-2c14a302838b0d0282be712c","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1437.7972257239153,"rawValue":1437.7972257239153,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=52","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-38b4491b6cd6cffa2898376c","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1425.36214121503,"rawValue":1425.36214121503,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=450","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-757b45f9b6e8cd04263e7dcf","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1428.1830694758214,"rawValue":1428.1830694758214,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=337","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-f5b465e139ae27b1b142cea3","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1475.6466788019832,"rawValue":1475.6466788019832,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=39","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-cacaf63e1ac9ad4c5afba893","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1475.6466788019832,"rawValue":1475.6466788019832,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=201","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-496d6bc49586f727f5aaa4f8","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1483.6255792907887,"rawValue":1483.6255792907887,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=319","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-035ce1e40fc92dac0bff4caa","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1464.0177570901815,"rawValue":1464.0177570901815,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=437","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-f564006ba6319898ac096e1f","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"minimax-m2.1-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1400.7095442722352,"rawValue":1400.7095442722352,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=348","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-28b503d2f0e3e059061c7048","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"minimax-m2.1-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1368.6020498541375,"rawValue":1368.6020498541375,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=468","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-b86ad80a5ff8ddb54f00ecec","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"minimax-m2.1-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1386.9898745078774,"rawValue":1386.9898745078774,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=72","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-1d6767c731acbf23882f2c3c","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"minimax-m2.1-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1386.9898745078774,"rawValue":1386.9898745078774,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=234","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-00fb96c01561c86605747f14","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"minimax-m2.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1414.5077245381826,"rawValue":1414.5077245381826,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=343","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-1e3ad7029c58ddc900da6e41","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"minimax-m2.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1372.5005082678872,"rawValue":1372.5005082678872,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=467","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-f389b66021c04ba0d77ef978","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"minimax-m2.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1383.8983643309668,"rawValue":1383.8983643309668,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=236","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-6696ec651ab7402f16badc1d","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"minimax-m2.5","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1383.8983643309668,"rawValue":1383.8983643309668,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=74","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-a50cf42bb1fd2a705644c343","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2.7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1387.7484436814018,"rawValue":1387.7484436814018,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=460","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-68205c581788076d0aab9a60","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2.7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1404.2259319703794,"rawValue":1404.2259319703794,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=346","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-1f31eed35149f88e89a51aad","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2.7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1398.13632454993,"rawValue":1398.13632454993,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=228","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-099e15a5f903b24ac39e1570","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"minimax-m2.7","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1398.13632454993,"rawValue":1398.13632454993,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=66","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-527644f2808f8d14a136bb34","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1472.4845412575187,"rawValue":1472.4845412575187,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=144","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-9755185a39fb818194d1bfb0","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1483.599854981654,"rawValue":1483.599854981654,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=431","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-6bb80820aa537a8defce02e4","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1487.6424274606477,"rawValue":1487.6424274606477,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=198","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-03eb8b537ee316cee8e4176e","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1487.6424274606477,"rawValue":1487.6424274606477,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=36","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-aeac3023cb9538106c240026","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1466.1016216771977,"rawValue":1466.1016216771977,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=325","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-3bd1acc0cb13686f7dc1ddfd","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"qwen3-coder-480b-a35b-instruct","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1256.494389060689,"rawValue":1256.494389060689,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=488","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-499cf2b6bb8bbc7ab2e91914","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"qwen3-coder-480b-a35b-instruct","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1273.2512988349522,"rawValue":1273.2512988349522,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=99","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-4592d2eb0ad3efe43bff2b14","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"qwen3-coder-480b-a35b-instruct","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1273.2512988349522,"rawValue":1273.2512988349522,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=261","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-562ac4c13c38e00720fe46f1","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"qwen3-coder-480b-a35b-instruct","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1286.321324993799,"rawValue":1286.321324993799,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=380","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-516bba21834f498cc65f7992","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1349.5224759985815,"rawValue":1349.5224759985815,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=477","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-aa9ebd87481100d7246c9eb2","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1357.6812637830733,"rawValue":1357.6812637830733,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=245","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-3895b7f168df80886b9e2ed3","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1357.6812637830733,"rawValue":1357.6812637830733,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=83","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-0e58c4db37d04a732397e272","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"qwen3.5-122b-a10b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1356.665847234471,"rawValue":1356.665847234471,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=366","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-55964976da0651b501b9889c","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1393.2231885877634,"rawValue":1393.2231885877634,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=353","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-a4d21243ddcd583b02bf0750","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1345.4015538321119,"rawValue":1345.4015538321119,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=479","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-a07f89c5807275152bb7eb05","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1357.4623741792195,"rawValue":1357.4623741792195,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=84","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-6157461d14b459de7375c628","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1357.4623741792195,"rawValue":1357.4623741792195,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=246","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-99cca0d7443c1b29b5a6169e","canonicalModelId":"qwen/qwen3.5-35b-a3b@current","modelName":"Qwen3.5 35B-A3B","modelRef":"qwen3.5-35b-a3b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1234.2886955279782,"rawValue":1234.2886955279782,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=491","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-44ddd6b24a71a662ebef7dba","canonicalModelId":"qwen/qwen3.5-35b-a3b@current","modelName":"Qwen3.5 35B-A3B","modelRef":"qwen3.5-35b-a3b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1298.2173308353517,"rawValue":1298.2173308353517,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=378","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-6414be94bd7f0e1d372e2e63","canonicalModelId":"qwen/qwen3.5-35b-a3b@current","modelName":"Qwen3.5 35B-A3B","modelRef":"qwen3.5-35b-a3b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1249.749391429547,"rawValue":1249.749391429547,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=104","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-5fa9ec92ebf66adfaf98cf4b","canonicalModelId":"qwen/qwen3.5-35b-a3b@current","modelName":"Qwen3.5 35B-A3B","modelRef":"qwen3.5-35b-a3b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1249.749391429547,"rawValue":1249.749391429547,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=266","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-9380e5836c8fd136f7a62caa","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1387.4312830150097,"rawValue":1387.4312830150097,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=461","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-f20f52f450efa767f3e869be","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1399.2275095999478,"rawValue":1399.2275095999478,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=65","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-e11a1442445cda15a3c6ab4d","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1399.2275095999478,"rawValue":1399.2275095999478,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=227","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-65de2e71f903c9e3f6fa45e7","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1410.9917137840002,"rawValue":1410.9917137840002,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=345","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-64b23242f467c220f5068a51","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1223.7875055486907,"rawValue":1223.7875055486907,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=494","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-09ae4e132218620b57e3dcad","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1293.103674188444,"rawValue":1293.103674188444,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=379","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-9e8a446d7a8461edde754493","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1238.1207909614184,"rawValue":1238.1207909614184,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=108","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-70c393818856d06ba07dc393","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1238.1207909614184,"rawValue":1238.1207909614184,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=270","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-4edfd67f5ea943226aeefda7","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1497.0504325208572,"rawValue":1497.0504325208572,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=315","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-e898cb707a547d7aba20dedd","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1478.8792606541383,"rawValue":1478.8792606541383,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=200","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-18bb38cde9ff712109d202da","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1478.8792606541383,"rawValue":1478.8792606541383,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=38","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-f50128c8cdccb93a5462411e","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1470.3582318681204,"rawValue":1470.3582318681204,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=436","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-b7de192f4c99085ff437af48","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1468.2702089284144,"rawValue":1468.2702089284144,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=145","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-9c02fbe78e8551c5343fc48d","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1453.822661483878,"rawValue":1453.822661483878,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=440","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-d8aa1c3d17c6a470ca0d4be5","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1459.7403782936024,"rawValue":1459.7403782936024,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=206","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-2fe89ebacedfa2f70d4136af","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1459.7403782936024,"rawValue":1459.7403782936024,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=44","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-1ae7a8d0afc0fb7595151d93","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1462.49644395851,"rawValue":1462.49644395851,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=327","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-15471b77846cc2aa0d6fc9f3","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max-20260517","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1514.03608266693,"rawValue":1514.03608266693,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=423","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-f61536ea371f53563acc3f0c","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max-20260517","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1516.6493621300187,"rawValue":1516.6493621300187,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=29","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-afc3735439dbc7e0c9985d3c","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max-20260517","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1516.6493621300187,"rawValue":1516.6493621300187,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=191","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-e931e4b11b60544554ef423f","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max-20260517","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1521.2481204137584,"rawValue":1521.2481204137584,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=310","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["curated_model_alias","human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-ec5dc3a85da0ece7a3388ee7","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1573.5016936795942,"rawValue":1573.5016936795942,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=124","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-cfe680009644a30ba26525f5","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1579.4229632505258,"rawValue":1579.4229632505258,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=288","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-b39eea8a1f2b9f23c0061e71","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1604.1093320820128,"rawValue":1604.1093320820128,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=405","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-633bd8e731b25e3c02716037","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1594.7775370883076,"rawValue":1594.7775370883076,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=8","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-5749757341d27bb136dd1bdd","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1594.7775370883076,"rawValue":1594.7775370883076,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=170","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-fecbf82fc2d37fe1b508116f","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1668.9390977471774,"rawValue":1668.9390977471774,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=164","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"elo"},{"id":"pub-b9033f9b36c7977546b02681","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1668.9390977471774,"rawValue":1668.9390977471774,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=2","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"overall","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"sourceMetricId":"elo"},{"id":"pub-9066c8932e8ce5f2f830383a","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1618.4401641653608,"rawValue":1618.4401641653608,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=120","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-25","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"image_to_webdev","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3,"sourceMetricId":"elo"},{"id":"pub-56ed1e99e1c853d6bf51c1ea","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1678.4351937220601,"rawValue":1678.4351937220601,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=400","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-react","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"sourceMetricId":"elo"},{"id":"pub-6e59474f534280e5fbbcfa26","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"arena-webdev","benchmarkName":"Arena WebDev","benchmarkVersionHint":"arena-webdev@latest","metricId":"arena_score_bt","value":1627.719515798759,"rawValue":1627.719515798759,"unit":"rating","evidenceLevel":"A","sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","sourcePageUrl":"https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset","sourceApiUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=285","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026-08-21","payloadSha256":"b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","protocol":{"arena_config":"webdev","category":"webdev-html","harness":"arena-human-preference","rating_method":"bradley-terry","split":"latest","subject_type":"system"},"harnessId":"arena-human-preference","qualityFlags":["human_preference_rating","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5,"sourceMetricId":"elo"},{"id":"pub-7f2979335561d82026ffa97e","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude-Haiku-4-5-20251001","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":68.7,"rawValue":"68.70%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=5;rank=6;model=Claude-Haiku-4-5-20251001 (FC);mode=FC","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"native_fc","calling_mode_label":"FC","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","missing_submetrics","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-226e95ae5dc15e984a9bae2f","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude-Haiku-4-5-20251001","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":25.26,"rawValue":"25.26%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=86;rank=87;model=Claude-Haiku-4-5-20251001 (Prompt);mode=Prompt","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"prompt","calling_mode_label":"Prompt","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-2226383995d7f954e72fa4f2","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"Claude-Opus-4-5-20251101","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":77.47,"rawValue":"77.47%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=0;rank=1;model=Claude-Opus-4-5-20251101 (FC);mode=FC","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"native_fc","calling_mode_label":"FC","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","missing_submetrics","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-2f6a6383db8cf0165ac4a8d9","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"Claude-Opus-4-5-20251101","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":33.47,"rawValue":"33.47%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=56;rank=57;model=Claude-Opus-4-5-20251101 (Prompt);mode=Prompt","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"prompt","calling_mode_label":"Prompt","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-91209b75e3bc183243a45082","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude-Sonnet-4-5-20250929","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":73.24,"rawValue":"73.24%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=1;rank=2;model=Claude-Sonnet-4-5-20250929 (FC);mode=FC","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"native_fc","calling_mode_label":"FC","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","missing_submetrics","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-bc605ce72442684036fcba27","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude-Sonnet-4-5-20250929","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":24.9,"rawValue":"24.90%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=88;rank=89;model=Claude-Sonnet-4-5-20250929 (Prompt);mode=Prompt","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"prompt","calling_mode_label":"Prompt","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-26ce2bdb7c58284b460e27cf","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":56.73,"rawValue":"56.73%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=13;rank=14;model=DeepSeek-V3.2-Exp (Prompt + Thinking);mode=Prompt + Thinking","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"prompt","calling_mode_label":"Prompt + Thinking","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":"Thinking"},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-5586ae213879dae6f344b7b9","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":54.12,"rawValue":"54.12%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=18;rank=19;model=DeepSeek-V3.2-Exp (FC);mode=FC","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"native_fc","calling_mode_label":"FC","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","missing_submetrics","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-b1b7deaae0ec48a3f84286b4","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5-mini-2025-08-07","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":55.46,"rawValue":"55.46%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=16;rank=17;model=GPT-5-mini-2025-08-07 (FC);mode=FC","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"native_fc","calling_mode_label":"FC","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","missing_submetrics","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-ca94f63940fd97e4c9f4d82b","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5-mini-2025-08-07","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":27.83,"rawValue":"27.83%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=76;rank=77;model=GPT-5-mini-2025-08-07 (Prompt);mode=Prompt","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"prompt","calling_mode_label":"Prompt","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-15e5d8399c6f1b82d96bec8c","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5-nano-2025-08-07","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":51.45,"rawValue":"51.45%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=23;rank=24;model=GPT-5-nano-2025-08-07 (FC);mode=FC","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"native_fc","calling_mode_label":"FC","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","missing_submetrics","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-66d7f2f32bf3bece91140771","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5-nano-2025-08-07","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":27.55,"rawValue":"27.55%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=78;rank=79;model=GPT-5-nano-2025-08-07 (Prompt);mode=Prompt","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"prompt","calling_mode_label":"Prompt","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-bdea4dfd0cff4904f2ac7e5a","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"GPT-5.2-2025-12-11","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":55.87,"rawValue":"55.87%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=15;rank=16;model=GPT-5.2-2025-12-11 (FC);mode=FC","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"native_fc","calling_mode_label":"FC","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","missing_submetrics","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-48710143ac56f4f2ae4452ee","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"GPT-5.2-2025-12-11","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":45.27,"rawValue":"45.27%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=37;rank=38;model=GPT-5.2-2025-12-11 (Prompt);mode=Prompt","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"prompt","calling_mode_label":"Prompt","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","curated_model_alias","missing_source_date","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-4ca0cc84048f1d321a8124b7","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"Gemini-2.5-Flash","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":56.24,"rawValue":"56.24%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=14;rank=15;model=Gemini-2.5-Flash (FC);mode=FC","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"native_fc","calling_mode_label":"FC","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","missing_source_date","missing_submetrics","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-8c36bfe4115a81beef1af1b0","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"Gemini-2.5-Flash","benchmarkId":"bfcl","benchmarkName":"Berkeley Function Calling Leaderboard","benchmarkVersion":"BFCL-V4","benchmarkVersionId":"bfcl@v4","metricId":"accuracy","value":50.9,"rawValue":"50.90%","unit":"percent","evidenceLevel":"A","sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourcePageUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceApiUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","sourceLocator":"csv-row=25;rank=26;model=Gemini-2.5-Flash (Prompt);mode=Prompt","retrievedAt":"2026-08-27T08:26:35Z","payloadSha256":"ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","protocol":{"benchmark_version":"BFCL-V4","benchmark_version_id":"bfcl@v4","calling_mode":"prompt","calling_mode_label":"Prompt","commit":"f7cf735","commit_url":"https://github.com/ShishirPatil/gorilla/commit/f7cf7359b7ac615a0b294831c5ba2bc95ee4a000","evaluator":"bfcl","evaluator_commit":"f7cf735","harness":"bfcl-eval","leaderboard_url":"https://gorilla.cs.berkeley.edu/leaderboard","subject_type":"system","variant":null},"harnessId":"bfcl-eval","qualityFlags":["benchmark_v4","missing_source_date","mode_specific","official_csv"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2,"evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard"},{"id":"pub-08f95142e630a2ecaab729f3","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-05-28","metricId":"Performance","value":804.12,"rawValue":"804.12","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=58;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1b596f19ea5bc0fe806bfe64","canonicalModelId":"deepseek/v3.1-terminus@2025-09-22","modelName":"DeepSeek V3.1 Terminus","modelRef":"DeepSeek-V3.1-Terminus","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-09-22","metricId":"Performance","value":745.17,"rawValue":"745.17","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=68;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e679a0da6147db5d7e002ab1","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"MiniMax-M2.1","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-12-23","metricId":"Performance","value":623.83,"rawValue":"623.83","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=80;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f5cd3d2031f255fbcaf84bf0","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-02-12","metricId":"Performance","value":618.17,"rawValue":"618.17","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=82;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f930522a1a8c26f73ae5f5a0","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMax-M2.7","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-03-18","metricId":"Performance","value":599.25,"rawValue":"599.25","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=83;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-aa44259f1d5c953f1a607afa","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-06-01","metricId":"Performance","value":640.02,"rawValue":"640.02","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=79;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-620bebf5b34932a105dbb5f9","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-07-31","metricId":"Performance","value":461.45,"rawValue":"461.45","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=92;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fb04f7a6d10a12372ef25b0a","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-06-09","metricId":"Performance","value":2041.3,"rawValue":"2041.3","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=4;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f2cd9e6ed2607e6c94c7a5d4","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_32K","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-10-15","metricId":"Performance","value":653.48,"rawValue":"653.48","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=78;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8331184009343cb360f560d6","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_16K","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-11-24","metricId":"Performance","value":1025.38,"rawValue":"1025.38","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=40;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a2ae1dd5a7b0057c189bde00","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-02-05","metricId":"Performance","value":996.5,"rawValue":"996.5","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=44;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-183796f64e699f8a1babe783","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-04-16","metricId":"Performance","value":1323.05,"rawValue":"1323.05","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=20;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-dd23e15b654e2c6d61a6037f","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-05-28","metricId":"Performance","value":1563.83,"rawValue":"1563.83","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=11;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7a60589d5bc488ad6344e1b5","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_none","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-05-28","metricId":"Performance","value":1411.83,"rawValue":"1411.83","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=16;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-628f5795d91a8d95ebc22b5d","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-07-24","metricId":"Performance","value":2164.57,"rawValue":"2164.57","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=3;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b389b3856bd32af288bfbdfb","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-09-29","metricId":"Performance","value":796.15,"rawValue":"796.15","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=61;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-28801e1a76bc7b0eceea7b56","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_medium","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-02-17","metricId":"Performance","value":1327.3,"rawValue":"1327.3","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=19;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d63d09f0680cff8abf3aa28e","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-06-30","metricId":"Performance","value":1463.12,"rawValue":"1463.12","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=15;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-cdfd4be1a5220fa234805f5b","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-07-31","metricId":"Performance","value":1306.08,"rawValue":"1306.08","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=23;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-296460808526f01e333b8304","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-04-24","metricId":"Performance","value":678.2,"rawValue":"678.2","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=73;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-1875f25bc38f30549bb69e49","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash_none","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-04-24","metricId":"Performance","value":324.98,"rawValue":"324.98","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=103;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-3bf4cef52b306abe0e0b18fb","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_max","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-08-13","metricId":"Performance","value":1403.15,"rawValue":"1403.15","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=17;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-14f794936dc5fb2b603d23dc","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-04-24","metricId":"Performance","value":1006.08,"rawValue":"1006.08","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=42;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-958f26b4a946d4720ea1be11","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_none","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-04-24","metricId":"Performance","value":521.67,"rawValue":"521.67","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=89;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ba86b78a196c859cd8066e85","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-06-17","metricId":"Performance","value":661.88,"rawValue":"661.88","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=76;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f1cd5de990ee912a25ba2a09","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro_32K","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-06-17","metricId":"Performance","value":785.52,"rawValue":"785.52","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=63;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3c9543e471986aa1dad45248","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-02-19","metricId":"Performance","value":1160.6,"rawValue":"1160.6","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=34;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2868c2583ec1b30bc82a12c2","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-07-21","metricId":"Performance","value":765.27,"rawValue":"765.27","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=66;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-66e06f179645e8ef8b0cb754","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-05-19","metricId":"Performance","value":911.02,"rawValue":"911.02","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=50;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d13525bb65774c0d038efbb5","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-07-21","metricId":"Performance","value":715.52,"rawValue":"715.52","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=71;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cb580ff1412486cc5c78ff78","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-08-13","metricId":"Performance","value":904.3,"rawValue":"904.3","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=51;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-43a7c18e6fcbd98305050800","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-12-22","metricId":"Performance","value":399.48,"rawValue":"399.48","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=95;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b8225008c03fd9c5ea71526a","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-02-11","metricId":"Performance","value":765.62,"rawValue":"765.62","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=65;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0801b90c67e84be86a1fd0af","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-04-07","metricId":"Performance","value":887.1,"rawValue":"887.1","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=53;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-59fea734a1f5ba152458c2c4","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-06-16","metricId":"Performance","value":1046.95,"rawValue":"1046.95","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=39;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-cbfab6bbff361085579f89c9","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-06-16","metricId":"Performance","value":1010.17,"rawValue":"1010.17","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=41;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-dd59b8065013c0cc8f017428","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-08-14","metricId":"Performance","value":1317.4,"rawValue":"1317.4","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=21;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-170514541587b3da76b11d49","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-08-07","metricId":"Performance","value":1162.45,"rawValue":"1162.45","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=33;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b5a9a27b44ef7d18f95de82c","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_minimal","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-08-07","metricId":"Performance","value":807.65,"rawValue":"807.65","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=57;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-119bc82c8c548ee4db7e4978","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-08-07","metricId":"Performance","value":799.77,"rawValue":"799.77","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=59;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-420059a80d99bf05b82c5304","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-08-07","metricId":"Performance","value":718.67,"rawValue":"718.67","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=70;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6c7e726f95f1c7ccb08d0c7f","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-11-13","metricId":"Performance","value":1192.15,"rawValue":"1192.15","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=29;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0d5f625fa7934080d9370263","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"gpt-5.1-codex","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-11-12","metricId":"Performance","value":1244.92,"rawValue":"1244.92","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=27;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-abc0d98495abc8bb7e4ffeee","canonicalModelId":"openai/gpt-5.1-codex-max@2025-11-19","modelName":"GPT-5.1 Codex Max","modelRef":"gpt-5.1-codex-max","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-11-19","metricId":"Performance","value":1208.83,"rawValue":"1208.83","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=28;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-861986abe64e9dd746408fca","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-12-11","metricId":"Performance","value":1293.55,"rawValue":"1293.55","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=25;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-765a0b6cbff72211b05f0908","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-12-11","metricId":"Performance","value":1249.83,"rawValue":"1249.83","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=26;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-3ee051ca13e374c249f70990","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2025-12-18","metricId":"Performance","value":1299.9,"rawValue":"1299.9","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=24;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1d1e8473403607f4293b4cc9","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex_xhigh","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-02-05","metricId":"Performance","value":1655.22,"rawValue":"1655.22","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=8;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-87b23ab32fa9ff65fa3be109","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-03-05","metricId":"Performance","value":1607.0,"rawValue":"1607.0","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=9;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-eafed19fdfecca2737dd73d1","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_medium","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-03-05","metricId":"Performance","value":1520.72,"rawValue":"1520.72","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=13;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2ce9d043d678154e297d2d77","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_none","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-03-05","metricId":"Performance","value":1086.03,"rawValue":"1086.03","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=38;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-d2d95193fcfa912ab56c73fd","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-03-17","metricId":"Performance","value":1188.58,"rawValue":"1188.58","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=31;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1ec17e7cfdd87e56b55bfd58","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-03-17","metricId":"Performance","value":1004.52,"rawValue":"1004.52","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=43;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e388e9153aebf944e13e95df","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_medium","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-04-23","metricId":"Performance","value":1589.38,"rawValue":"1589.38","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=10;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-27487a09e289be8dc4b40a56","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_none","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-04-23","metricId":"Performance","value":1127.58,"rawValue":"1127.58","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=36;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-bae240ff8814622c718f7c79","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-04-23","metricId":"Performance","value":1942.97,"rawValue":"1942.97","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=6;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-081230318beba6136e374c92","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-07-09","metricId":"Performance","value":1667.4,"rawValue":"1667.4","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=7;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8c9e70fb91a898aa6102d52a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-07-09","metricId":"Performance","value":2176.88,"rawValue":"2176.88","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=2;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e9418e1af546fa1a58b2d564","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-07-09","metricId":"Performance","value":1951.42,"rawValue":"1951.42","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=5;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8c35b155c6761f7cceb4ab87","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-07-08","metricId":"Performance","value":1309.28,"rawValue":"1309.28","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=22;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-58c4a65def19878e4e832b4b","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_xhigh","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-08-12","metricId":"Performance","value":1508.08,"rawValue":"1508.08","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=14;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5f4d900ae4977e97e1068000","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-01-27","metricId":"Performance","value":821.65,"rawValue":"821.65","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=56;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c8296981921bc361026c3eb3","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-04-20","metricId":"Performance","value":1092.67,"rawValue":"1092.67","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=37;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fa861330e75c2fa6bcc14c07","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-06-12","metricId":"Performance","value":886.23,"rawValue":"886.23","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=54;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8decfe7c72269457ea7bc129","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-07-16","metricId":"Performance","value":1524.47,"rawValue":"1524.47","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=12;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1eb273294e8cd0b336a4f9c4","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash","benchmarkId":"epoch-ale_bench_external","metricId":"Performance","value":737.95,"rawValue":"737.95","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=69;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3b5b8ae3a6689bb341ce1c44","canonicalModelId":"xiaomi/mimo-v2-pro@current","modelName":"MiMo V2 Pro","modelRef":"mimo-v2-pro","benchmarkId":"epoch-ale_bench_external","metricId":"Performance","value":785.17,"rawValue":"785.17","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=64;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3a5431afe458a6a55601760c","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"epoch-ale_bench_external","metricId":"Performance","value":513.95,"rawValue":"513.95","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=90;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-487e23487eb0211e4c172a58","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"epoch-ale_bench_external","metricId":"Performance","value":899.8,"rawValue":"899.8","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=52;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4bf6951622c9f7c88ea95cd1","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27B","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-02-24","metricId":"Performance","value":349.45,"rawValue":"349.45","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=98;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c184cb5c6866aeeef28c9685","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-02-25","metricId":"Performance","value":221.8,"rawValue":"221.8","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=107;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-eda775488ae4ac62e384d36a","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-02-16","metricId":"Performance","value":621.92,"rawValue":"621.92","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=81;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6b73981ff3aa527b3270bb59","canonicalModelId":"qwen/qwen3.6-flash@2026-04-27","modelName":"Qwen3.6 Flash","modelRef":"qwen3.6-flash","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-04-27","metricId":"Performance","value":326.4,"rawValue":"326.4","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=101;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e22b349a91a118c65ca4ecc9","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-03-31","metricId":"Performance","value":670.15,"rawValue":"670.15","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=75;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4e59c2fe2e06a031eb3ddecc","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"epoch-ale_bench_external","benchmarkVersion":"2026-05-19","metricId":"Performance","value":1189.4,"rawValue":"1189.4","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=30;column=Performance","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-34d6ad7daeb2cda6eba2f07a","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"epoch-algotune_external","benchmarkVersion":"2025-07-31","metricId":"Score","value":1.44,"rawValue":"1.44","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=14;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-53b8e0181f0f4ebe8a15654e","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_unknown","benchmarkId":"epoch-algotune_external","benchmarkVersion":"2025-05-22","metricId":"Score","value":1.33,"rawValue":"1.33","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=18;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2b0532399b00694580a43732","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-algotune_external","benchmarkVersion":"2025-11-24","metricId":"Score","value":1.77,"rawValue":"1.77","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=6;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c0318621c099e83bc42fe7b6","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_high","benchmarkId":"epoch-algotune_external","benchmarkVersion":"2026-02-05","metricId":"Score","value":1.47,"rawValue":"1.47","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=13;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-542ccd20a1f93474d6cc8d15","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-algotune_external","benchmarkVersion":"2025-09-29","metricId":"Score","value":1.52,"rawValue":"1.52","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=11;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-621d5ef0499060ba2ecd0668","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-algotune_external","benchmarkVersion":"2025-06-17","metricId":"Score","value":1.51,"rawValue":"1.51","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=12;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-16e40cfb28640bfa1e50ad29","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-algotune_external","benchmarkVersion":"2026-02-19","metricId":"Score","value":2.02,"rawValue":"2.02","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=3;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-47b5ab1ff7400f6c8ac2ae3d","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-algotune_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":1.67,"rawValue":"1.67","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=9;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-70f182c1d468d887694f64b3","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-algotune_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":1.38,"rawValue":"1.38","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=16;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-699e423858371a1a01f008b9","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-algotune_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":2.05,"rawValue":"2.05","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=2;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-72c68628ba670407c9e94ce2","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"epoch-algotune_external","benchmarkVersion":"2026-03-05","metricId":"Score","value":1.85,"rawValue":"1.85","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=4;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b47d1b696cda8796a1f80b01","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-02-12","metricId":"Pass@1 score","value":6.200000000000001,"rawValue":"0.062000000000000006","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=56;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-be6749e8bf13b8e9b97142ae","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-06-09","metricId":"Pass@1 score","value":45.0,"rawValue":"0.45","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=2;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-76c5f3957158ceda041f0e7e","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-10-15","metricId":"Pass@1 score","value":8.9,"rawValue":"0.08900000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=51;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-25aa03c73e74364f3458c7aa","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-11-24","metricId":"Pass@1 score","value":20.7,"rawValue":"0.207","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=34;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bd317b900bdc406de9d23557","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-02-05","metricId":"Pass@1 score","value":32.1,"rawValue":"0.321","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=23;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8f8eb966756fb3bf96f86501","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-02-05","metricId":"Pass@1 score","value":32.4,"rawValue":"0.324","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=22;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-2ee9e8ee5dda8c499ac63c98","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-04-16","metricId":"Pass@1 score","value":33.900000000000006,"rawValue":"0.339","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=19;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e8ccf2fff85b759c940d33e3","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-05-28","metricId":"Pass@1 score","value":42.5,"rawValue":"0.425","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=4;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8f88d8806baf4ac06df5f44f","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-07-24","metricId":"Pass@1 score","value":43.5,"rawValue":"0.435","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=3;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3a8111cb9264d11d9d5bf8f3","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_high","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-02-17","metricId":"Pass@1 score","value":23.7,"rawValue":"0.237","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=30;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-90e256cc9318eaaf06bf470c","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-06-30","metricId":"Pass@1 score","value":32.5,"rawValue":"0.325","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=21;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-46deb19c23cafd2a4124b65b","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek/deepseek-v3.2","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-12-01","metricId":"Pass@1 score","value":7.000000000000001,"rawValue":"0.07","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=54;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-219dceba4276a18945482690","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-06-17","metricId":"Pass@1 score","value":1.8000000000000003,"rawValue":"0.018000000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=61;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d2fd7affb7d4fe8822fa5214","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-06-17","metricId":"Pass@1 score","value":6.6000000000000005,"rawValue":"0.066","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=55;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a1d2b7557e476c4b734d5cbf","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-02-19","metricId":"Pass@1 score","value":33.5,"rawValue":"0.335","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=20;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-51e1e65f3dc74278a2bb550a","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-12-22","metricId":"Pass@1 score","value":8.7,"rawValue":"0.087","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=53;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-89fa49a1b66404c7d6dea15d","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-02-11","metricId":"Pass@1 score","value":17.2,"rawValue":"0.172","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=42;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-92b76508abeb5b0774bb57c8","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-06-16","metricId":"Pass@1 score","value":35.6,"rawValue":"0.35600000000000004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=15;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-591469dcb64a4ef817a738cf","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-08-07","metricId":"Pass@1 score","value":18.3,"rawValue":"0.18300000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=37;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-8b880bba849df0b39e58cdea","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-08-07","metricId":"Pass@1 score","value":18.3,"rawValue":"0.18300000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=38;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cd59b5c08ddad55696875b0e","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-11-13","metricId":"Pass@1 score","value":17.5,"rawValue":"0.17500000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=39;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ba106e63b4a62fba4148fba2","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-11-13","metricId":"Pass@1 score","value":17.5,"rawValue":"0.17500000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=40;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-0279c53be397a5efae1c95eb","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"gpt-5.1-codex","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-11-12","metricId":"Pass@1 score","value":20.7,"rawValue":"0.207","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=33;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c2738290a213b5ab4afb0b4e","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-12-11","metricId":"Pass@1 score","value":23.0,"rawValue":"0.23","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=31;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-72f838f3ed5af7a28fb6d6ff","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-12-11","metricId":"Pass@1 score","value":23.0,"rawValue":"0.23","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=32;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-62a4e5b933f21e79e46af18c","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-12-11","metricId":"Pass@1 score","value":34.4,"rawValue":"0.344","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=17;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-3b4153f33032104cdf0ed37a","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-12-18","metricId":"Pass@1 score","value":27.6,"rawValue":"0.276","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=26;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fc4b3f42892789c139ad94c0","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-02-05","metricId":"Pass@1 score","value":31.8,"rawValue":"0.318","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=24;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8cd169b0411110ac6ca1623b","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-03-05","metricId":"Pass@1 score","value":34.9,"rawValue":"0.349","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=16;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7eea6b9cf09843f9403c7a5a","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-03-05","metricId":"Pass@1 score","value":36.0,"rawValue":"0.36","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=14;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f7a54da3c445064fda56ae50","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-03-17","metricId":"Pass@1 score","value":24.6,"rawValue":"0.24600000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=28;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-aa2e8459fe2c1bbca722faff","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-03-17","metricId":"Pass@1 score","value":16.9,"rawValue":"0.16899999999999998","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=43;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-31166d2a718a0ce358877339","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-04-23","metricId":"Pass@1 score","value":38.5,"rawValue":"0.385","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=11;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f4b3a113059f4a8bde205769","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-04-23","metricId":"Pass@1 score","value":38.5,"rawValue":"0.385","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=10;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a8961600fe2d45c34365b7ad","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1 score","value":39.900000000000006,"rawValue":"0.399","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=8;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-a0662b31c62c3fee1a36980f","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_promax","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1 score","value":40.0,"rawValue":"0.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=7;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-db526d75da164588d62149fd","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1 score","value":37.7,"rawValue":"0.37700000000000006","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=12;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f566e7de5ff2d452b5dce791","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1 score","value":37.7,"rawValue":"0.37700000000000006","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=13;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-83aca449f74acdb850a4cec2","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-07-08","metricId":"Pass@1 score","value":34.2,"rawValue":"0.342","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=18;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f025d0651b750f7042a21075","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-08-12","metricId":"Pass@1 score","value":41.2,"rawValue":"0.41200000000000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=6;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-17a772396638a574f853c02a","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-01-27","metricId":"Pass@1 score","value":14.400000000000002,"rawValue":"0.14400000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=45;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-73e013504984321c2b94d76d","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-04-20","metricId":"Pass@1 score","value":18.9,"rawValue":"0.189","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=36;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-150873dde4b2fded68d184df","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-06-12","metricId":"Pass@1 score","value":27.6,"rawValue":"0.276","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=27;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4e489572efc37fffeb3bb4dd","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_unknown","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-07-16","metricId":"Pass@1 score","value":39.3,"rawValue":"0.39299999999999996","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=9;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-31f05c3f601c000748fbe460","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2026-02-16","metricId":"Pass@1 score","value":13.600000000000001,"rawValue":"0.136","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=46;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c63bf258832767ea3183a830","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/glm-4.7","benchmarkId":"epoch-apex_agents_external","benchmarkVersion":"2025-12-22","metricId":"Pass@1 score","value":8.7,"rawValue":"0.087","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=52;column=Pass@1 score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f16473033ab40169ff7469ef","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-05-28","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0112,"rawValue":"0.0112","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=183;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-07950142bbfd0e668bd6b767","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-02-12","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0486,"rawValue":"0.0486","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=134;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-30cccee9f26dd193839918b2","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-06-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.875,"rawValue":"0.875","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=10;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-d3a4d07d2c7265d91f4f8743","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-06-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.7680555555555555,"rawValue":"0.7680555555555555","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=26;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-cd86b0c38c17b14273d6138c","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-06-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8916666666666667,"rawValue":"0.8916666666666667","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=6;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-945cab112d06431f79571f72","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-06-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8250000000000001,"rawValue":"0.8250000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=23;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-f0378b407623ccdf5868d0b9","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-06-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8833333333333333,"rawValue":"0.8833333333333333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=7;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7a0eca5252f5d51481006f19","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-10-15","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0125,"rawValue":"0.0125","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=180;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-0c4ba39afe9a9368989bb1e0","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_16K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-10-15","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0278,"rawValue":"0.0278","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=150;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-05ba3c2ff8efd1b562087d26","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_1K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-10-15","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0125,"rawValue":"0.0125","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=182;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-4b4b41a93933fdf6a8c5793b","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_32K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-10-15","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0403,"rawValue":"0.0403","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=141;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-d40233fbbe3383aa912ce96a","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_8K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-10-15","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0167,"rawValue":"0.0167","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=171;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0fdc8ec58bfc123d8e361935","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.012700000000000001,"rawValue":"0.012700000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=178;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-2f9ca887507818ee273f219e","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_16K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0861,"rawValue":"0.0861","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=114;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-610993504638601fe7c6a0aa","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_1K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0,"rawValue":"0.0","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=208;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-7d4debf90b688f510702ca44","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_8K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0452,"rawValue":"0.0452","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=136;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9a3e2545d19ab4093ee68015","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0778,"rawValue":"0.0778","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=115;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-737a63b75a5cf82b6a0a59fc","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_16K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.2278,"rawValue":"0.2278","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=95;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-2ca3452da8e18de92b16c43f","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_64K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.3764,"rawValue":"0.3764","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=82;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-856cb23a4c100f9d9ce4d7c9","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_8K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.1389,"rawValue":"0.1389","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=103;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e0bd296f88d001b9819aa32c","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_120K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6458,"rawValue":"0.6458","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=49;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c561c0f3fe8ddc8465f7f6cb","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_120K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6625,"rawValue":"0.6625","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=46;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-87d5d8ba27dfae1061ec1b8c","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_120K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6875,"rawValue":"0.6875","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=38;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-35e612f8cf086a3e40628512","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_120K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6917,"rawValue":"0.6917","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=37;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-cc79d1cf35bef54be8c3dbcb","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.683,"rawValue":"0.683","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=40;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-899241c7fe6c5765d01120db","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6833,"rawValue":"0.6833","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=39;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-a2e88669d9c12d19c65b8cde","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6208,"rawValue":"0.6208","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=52;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-22810510b3411c0b1d48a4d0","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.758,"rawValue":"0.758","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=28;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-0743eee583f9307924d6f7cc","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.7583,"rawValue":"0.7583","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=27;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-92ba8317800a6f211d0185b8","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.675,"rawValue":"0.675","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=42;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b3d47a44f265f31167d283e3","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-05-28","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.7208,"rawValue":"0.7208","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=34;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-e12efa80d9fe60bd97571c16","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-05-28","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6222,"rawValue":"0.6222","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=51;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-93b26600e41a35f91ebd74d8","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-05-28","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.7167,"rawValue":"0.7167","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=35;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-d3fcb1763098984d825cb5ab","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-24","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8833,"rawValue":"0.8833","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=9;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-79846a7da0fa5a448a0a2245","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-24","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8833,"rawValue":"0.8833","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=8;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-32d02a391188d95dbd3f5beb","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-24","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.9042,"rawValue":"0.9042","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=3;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-5c41c632082691755df29309","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0375,"rawValue":"0.0375","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=144;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-bcead6c91dd54a443f135bf1","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_16K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0694,"rawValue":"0.0694","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=118;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e4743fea940171b090960325","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_1K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.058300000000000005,"rawValue":"0.058300000000000005","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=125;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-430db64063804a25963a128b","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.1361,"rawValue":"0.1361","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=105;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-559c424fcb9338350174ba73","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_8K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0694,"rawValue":"0.0694","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=119;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-d34801ab03ea3fa4089df112","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-02-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6042,"rawValue":"0.6042","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=56;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c21286b292a1f0a0fe8a56b8","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-02-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5833,"rawValue":"0.5833","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=61;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-a2dc344ae4976feb4dbfa804","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-31","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5597222222222221,"rawValue":"0.5597222222222221","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=63;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-dafdf6f5778b04a369deff32","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-31","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.4597222222222222,"rawValue":"0.4597222222222222","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=76;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-074951183441b91a529752c1","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-31","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6138888888888888,"rawValue":"0.6138888888888888","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=53;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-14bcd2e446aee776121047ac","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_none","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-31","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.020833333333333332,"rawValue":"0.020833333333333332","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=158;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-e076f37f945447d1794da13f","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-08-13","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5972222222222221,"rawValue":"0.5972222222222221","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=59;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b33f164c45ceb105c90a6c9b","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-08-13","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5625,"rawValue":"0.5625","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=62;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ab467e0edcf09ee76c04110f","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-08-13","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6125,"rawValue":"0.6125","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=55;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-8dc985c426b584ce1ff122f2","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_none","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-08-13","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.008333333333333333,"rawValue":"0.008333333333333333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=189;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-48116e393e548f09a9c944a4","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek/deepseek-v3.2","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-12-01","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0403,"rawValue":"0.0403","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=142;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-46c0062a2391ef9a6a815574","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro_32K","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-06-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0486,"rawValue":"0.0486","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=133;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a7962a79f31391d8c3216f27","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-02-19","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.7708,"rawValue":"0.7708","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=25;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-507c5930078d095c420f0c91","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-02-19","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.771,"rawValue":"0.771","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=24;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-625595816df4a1bf265bf7c4","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.10277777777777779,"rawValue":"0.10277777777777779","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=110;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-a9beed91931b0d7197029832","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.015277777777777777,"rawValue":"0.015277777777777777","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=173;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-3dcdcc989c7515b59d7c4863","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.05277777777777778,"rawValue":"0.05277777777777778","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=129;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-f2ac146241074ad55bd6e620","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_minimal","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.008333333333333333,"rawValue":"0.008333333333333333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=188;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-250638ff4fe083bebb87911d","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.7208,"rawValue":"0.7208","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=33;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c95b4a32a6dbbda19d1dccfd","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_minimal","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0889,"rawValue":"0.0889","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=113;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7670b2d188872692fdbe9dda","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6041666666666666,"rawValue":"0.6041666666666666","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=58;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-95afd5c84e27c5e7ec473b9f","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.30416666666666664,"rawValue":"0.30416666666666664","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=90;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-7b6c1f094530f9bb9d7ea3cc","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5041666666666667,"rawValue":"0.5041666666666667","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=73;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ba5ca099786e90833be4c7c1","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_minimal","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.02638888888888889,"rawValue":"0.02638888888888889","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=151;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6ebd601f300b0f87f0c54913","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-08-13","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8458333333333333,"rawValue":"0.8458333333333333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=14;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-a4e1991fa75ccc4c5b2a6b69","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-08-13","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5291666666666667,"rawValue":"0.5291666666666667","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=67;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-222a754e2b55e577f7eb8153","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-08-13","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6375,"rawValue":"0.6375","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=50;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-5298f738c3120aa41ba8c520","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-02-11","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0486,"rawValue":"0.0486","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=135;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3aa27780a2e2611e54e56e9a","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_unknown","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-06-16","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.22775000000000004,"rawValue":"0.22775000000000004","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=96;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-74a006add7521303306e09be","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0986,"rawValue":"0.0986","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=111;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b8bc03bbdde281f92fd7920f","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0194,"rawValue":"0.0194","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=165;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4f6a3ec24d0de6f2338a320a","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0194,"rawValue":"0.0194","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=164;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-93fb295a2da8a38cdf45037f","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0749,"rawValue":"0.0749","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=116;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-6f38bc0c2e9e20e880bccdfd","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_minimal","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0,"rawValue":"0.0","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=212;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-cacf11962a7c250daecf90e2","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0444,"rawValue":"0.0444","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=137;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-0ca85eeb4ae4eba38137c748","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0083,"rawValue":"0.0083","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=192;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-19b407e1fb1c9d314dfa840d","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0403,"rawValue":"0.0403","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=140;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-e2456e62f7c381b3794aa7a8","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_minimal","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0167,"rawValue":"0.0167","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=170;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-316d2320396a1eb5fd0ecd75","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.026100000000000005,"rawValue":"0.026100000000000005","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=152;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-b54876a9f0e33926f8340faa","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0,"rawValue":"0.0","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=213;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-73bf205c4290c5d8989d5de2","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0088,"rawValue":"0.0088","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=186;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-6d80c09544c57188156bebad","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_minimal","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0,"rawValue":"0.0","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=215;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-fbeaf4858ffb07bfebc01adc","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.1764,"rawValue":"0.1764","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=101;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-80f9b7e684bbf904fd5b83e6","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0194,"rawValue":"0.0194","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=166;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-2a0a3ca2eaeda7f4502d0b02","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0653,"rawValue":"0.0653","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=121;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-a7fa6d643cf441a5e6d1f81e","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_none","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0042,"rawValue":"0.0042","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=198;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f7fc25ec5a6b446c1024c991","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.4333,"rawValue":"0.4333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=78;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6e3f5ccb5a12db2d923e738e","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0972,"rawValue":"0.0972","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=112;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-6fe2de3e1a14639e6a79bd43","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.2667,"rawValue":"0.2667","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=94;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-70e898046700010c584631f3","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0083,"rawValue":"0.0083","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=191;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-88ec01be3dfcecd8d4d4bbe1","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5291,"rawValue":"0.5291","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=68;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-38e16992d5909881ede4b829","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5291,"rawValue":"0.5291","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=69;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-32441847c6e40713074144e5","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5291,"rawValue":"0.5291","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=70;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-bd78ee464d8c92e0c1233ccf","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5416,"rawValue":"0.5416","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=66;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d269f45b4160ff71d9c375e8","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.3847,"rawValue":"0.3847","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=81;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d39003ad4df5892e677ec7e1","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.675,"rawValue":"0.675","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=41;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6a9b8aa67557eea4f4c2b301","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.2917,"rawValue":"0.2917","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=92;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-5ef89146f2618d4e1a7eca39","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5542,"rawValue":"0.5542","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=64;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-86595bb0b05e946c711a757a","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.7395,"rawValue":"0.7395","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=32;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-454e35a82a355920304cb4dc","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.7395,"rawValue":"0.7395","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=31;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-b8d50051c9c15365645c5bbc","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.1319,"rawValue":"0.1319","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=106;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-5db9e7bfd9f83e2559989b3e","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0111,"rawValue":"0.0111","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=184;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-b4994dfaf7c762f755d54bee","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0444,"rawValue":"0.0444","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=138;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-dcca6adce75c01c6f8521c67","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.189,"rawValue":"0.189","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=97;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-67b35b90c449ecef437a7ee1","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.189,"rawValue":"0.189","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=98;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-fda373a09a8a41a004946e85","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0361,"rawValue":"0.0361","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=145;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e92287199b55b1c6ef451ed7","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.015300000000000001,"rawValue":"0.015300000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=172;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-1b66738ddc15ac82e625ad17","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0194,"rawValue":"0.0194","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=167;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-639555bdd7bf4e20e2845df4","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0569,"rawValue":"0.0569","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=126;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-5a0809e9d142650447128f7c","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0569,"rawValue":"0.0569","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=127;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-5cbad0c25dcb8aa9efe15593","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8333,"rawValue":"0.8333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=20;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-087f97615c253901f1f74a7d","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8333,"rawValue":"0.8333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=22;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-6634c4224c0492cfe66d4593","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-23","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8458,"rawValue":"0.8458","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=16;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-cd281010a1526e1febfc0853","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-23","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8416,"rawValue":"0.8416","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=17;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-99d6a35110014abea727d0c0","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-23","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8416,"rawValue":"0.8416","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=18;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b99bf17e66ddefa4c7df4dc0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-23","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8333,"rawValue":"0.8333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=21;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-08ff65141718909cc5042271","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-23","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.3333,"rawValue":"0.3333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=86;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-afdbccda48671a06ea937a77","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-23","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.7042,"rawValue":"0.7042","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=36;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-81af8866043a774d13723f5f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-23","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.85,"rawValue":"0.85","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=13;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-0774ba0b56b41d6fdf421ac1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-04-23","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.85,"rawValue":"0.85","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=12;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-cdf490e39b67653813e1de8b","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.2931,"rawValue":"0.2931","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=91;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d8d6aa2443f2a5dc6f338569","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.05140000000000001,"rawValue":"0.05140000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=130;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-556d2ae836e1d5480843a754","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5954,"rawValue":"0.5954","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=60;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-340f4b8142d56334e84071bb","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.0736,"rawValue":"0.0736","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=117;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-9f245496f5b36e7c1876e186","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.4764,"rawValue":"0.4764","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=74;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-1d0c8f57aa5dcadc8030eea0","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.4764,"rawValue":"0.4764","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=75;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-f86a314466c739d09e1e23b3","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.8542000000000001,"rawValue":"0.8542000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=11;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c85af1de8b533c4e47ce9e01","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.425,"rawValue":"0.425","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=79;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-80c16492719fef5a1faa88f4","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.925,"rawValue":"0.925","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=2;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-e281b08c8eec4426044e1662","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6708,"rawValue":"0.6708","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=44;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b0463dd89f46f2c47c1c2988","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.9,"rawValue":"0.9","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=4;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-5ab2d1c7bd81afefe902c239","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.9,"rawValue":"0.9","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=5;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-4b8396a39876daf7c3d22c89","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6708,"rawValue":"0.6708","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=45;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-e4a426f37c122049c079771d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.1875,"rawValue":"0.1875","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=99;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-87714bab4a8345de2639065a","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.839,"rawValue":"0.839","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=19;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-f3b45746ffaec465e90e3e12","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.375,"rawValue":"0.375","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=83;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b69004f92f99f90197722f2b","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.7417,"rawValue":"0.7417","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=30;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-04c29303869037dc4f9c6ce2","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.7417,"rawValue":"0.7417","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=29;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-d689582f4bac5287b4905a86","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2025-09-19","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.05280000000000001,"rawValue":"0.05280000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=128;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-641c96d7808ebcc0c041570f","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-08","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5264,"rawValue":"0.5264","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=71;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-9839aa1fc8d9a30dee27532f","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-08","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.3306,"rawValue":"0.3306","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=87;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1d04d0287c4a6c1cc2845abf","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-08","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.5264,"rawValue":"0.5264","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=72;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-f012c23378581827b2b49997","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-08-12","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6513888888888889,"rawValue":"0.6513888888888889","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=48;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9b42ac8c05233718ed4b65ac","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-08-12","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.2763888888888889,"rawValue":"0.2763888888888889","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=93;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-b7566b7519ff656ea3234977","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_medium","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-08-12","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6125,"rawValue":"0.6125","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=54;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-297ab1f168070b64deb3526e","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_xhigh","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-08-12","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6708333333333333,"rawValue":"0.6708333333333333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=43;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-76613d0e03a094a49630f452","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-01-27","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.11810000000000001,"rawValue":"0.11810000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=109;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3ab394af3fec3d18390e7d1a","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_high","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-16","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.55,"rawValue":"0.55","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=65;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1ab4fa9b4477e4b4a778f1f5","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_low","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-16","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.12361111111111112,"rawValue":"0.12361111111111112","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=108;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-21296d7ab4a821123bbe468e","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-arc_agi_2_external","benchmarkName":"ARC-AGI-2 (Epoch external aggregation)","benchmarkVersion":"2026-07-16","benchmarkVersionHint":"epoch-arc_agi_2_external@rolling","metricId":"Score","value":0.6041666666666666,"rawValue":"0.6041666666666666","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=57;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-e589c92938cc1e956124a029","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-05-28","metricId":"Score","value":0.2121,"rawValue":"0.2121","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=195;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-39ebf715d18bab5dda0eddd7","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-02-12","metricId":"Score","value":0.6367,"rawValue":"0.6367","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=117;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1bab30ce164348b8abd5757a","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.9550000000000001,"rawValue":"0.9550000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=16;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-3810149a760711b7ea894842","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.905,"rawValue":"0.905","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=54;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-b7c66d5b7e242cda97717361","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.985,"rawValue":"0.985","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=3;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-96efbb27623b6ecf546984b5","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.925,"rawValue":"0.925","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=39;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b84d9027a08b117781e1f12c","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.985,"rawValue":"0.985","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=2;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-acef982b7100ca5a987f7f4b","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-10-15","metricId":"Score","value":0.1433,"rawValue":"0.1433","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=209;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-54fb54c48737804a4c90999d","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_16K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-10-15","metricId":"Score","value":0.3733000000000001,"rawValue":"0.3733000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=158;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-b61b691e21394641ddfed2ae","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_1K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-10-15","metricId":"Score","value":0.16830000000000003,"rawValue":"0.16830000000000003","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=201;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-33cf92dd8971bf5b2710f06b","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_32K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-10-15","metricId":"Score","value":0.4767,"rawValue":"0.4767","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=144;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-c9a56dc82da84c745dc43385","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_8K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-10-15","metricId":"Score","value":0.255,"rawValue":"0.255","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=188;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4ecc64b43df2855bc31db99d","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-05-22","metricId":"Score","value":0.225,"rawValue":"0.225","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=191;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-74f8f4d8faf620db630d8ea0","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_16K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-05-22","metricId":"Score","value":0.3567,"rawValue":"0.3567","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=160;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-3885aa026de0dbe5cde2c53c","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_1K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-05-22","metricId":"Score","value":0.27,"rawValue":"0.27","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=184;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-dea6cb329a3a13ac470d97eb","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_8K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-05-22","metricId":"Score","value":0.3067,"rawValue":"0.3067","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=176;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-870eb2d6d76c2e165f852d06","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-24","metricId":"Score","value":0.4,"rawValue":"0.4","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=155;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-53dee251d96b462c75a6d5ef","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_16K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-24","metricId":"Score","value":0.72,"rawValue":"0.72","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=107;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-e913f1c03bb542667801ecdc","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_32K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-24","metricId":"Score","value":0.7583,"rawValue":"0.7583","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=98;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a9222f7f35638951287ff069","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_32K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-24","metricId":"Score","value":0.758,"rawValue":"0.758","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=99;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-c2e10c33b3163d925347672b","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_64K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-24","metricId":"Score","value":0.8,"rawValue":"0.8","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=88;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-acc1052941321281ec1f7223","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_64K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-24","metricId":"Score","value":0.8,"rawValue":"0.8","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=87;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-8a8464267ad81a7375478ab9","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_8K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-24","metricId":"Score","value":0.5867,"rawValue":"0.5867","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=124;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-eb80f723deda646fa51f81e7","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_120K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-02-05","metricId":"Score","value":0.9400000000000001,"rawValue":"0.9400000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=27;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-beea5e1b24c2ed1ac3d00bb4","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_120K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-02-05","metricId":"Score","value":0.93,"rawValue":"0.93","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=34;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-8cdd1259ecc29c4c4a60a658","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_120K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-02-05","metricId":"Score","value":0.92,"rawValue":"0.92","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=44;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-265d12139236d5d7f8f013b0","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_120K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-02-05","metricId":"Score","value":0.86,"rawValue":"0.86","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=78;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-20bb7638d96f5c54d65932e8","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_120K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-02-05","metricId":"Score","value":0.94,"rawValue":"0.94","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=29;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-6bcce2bff4eb81988f03bb3e","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.935,"rawValue":"0.935","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=32;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-234d5b0b60c1d2a0b7b49031","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.935,"rawValue":"0.935","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=33;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-4e2012a14a7189e3a923ce2f","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.91,"rawValue":"0.91","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=49;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-c9486ca69bc99536e46f7312","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.92,"rawValue":"0.92","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=41;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6f1b4a693c3728e9ea531fa7","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.92,"rawValue":"0.92","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=45;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b1977b734acbb0ab2032ff16","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.91,"rawValue":"0.91","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=50;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b86081ef4df72c7080073c77","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.92,"rawValue":"0.92","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=42;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-e3e06a390e56d3cb96d45f8c","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.88,"rawValue":"0.88","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=59;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ba0b957ea1ba7b09ad4c0a16","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.925,"rawValue":"0.925","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=37;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-5967975c4f5abbbe150b2475","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.915,"rawValue":"0.915","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=46;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-38ab2ce1db079da5025de074","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.975,"rawValue":"0.975","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=7;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-db1e418e6251a87bfe7e504b","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.975,"rawValue":"0.975","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=6;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-407fd6ef17bc4f1ed36498a1","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.975,"rawValue":"0.975","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=8;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-941e9e394afdb2f8f697de45","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-09-29","metricId":"Score","value":0.255,"rawValue":"0.255","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=187;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-6b273019b27620906ba575b4","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_16K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-09-29","metricId":"Score","value":0.4833,"rawValue":"0.4833","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=143;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-16c14a5808c2dd346d30be7c","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_1K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-09-29","metricId":"Score","value":0.31,"rawValue":"0.31","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=174;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-d6bc1fe895607b83484c9fb4","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-09-29","metricId":"Score","value":0.6367,"rawValue":"0.6367","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=116;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-035cc1c66b5b9f57554fcf7b","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_8K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-09-29","metricId":"Score","value":0.465,"rawValue":"0.465","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=145;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-fadbfe5799bbbba64baefc8e","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-02-17","metricId":"Score","value":0.865,"rawValue":"0.865","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=72;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-047929dd5fc0c592e7c72e86","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-02-17","metricId":"Score","value":0.86,"rawValue":"0.86","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=77;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-bd18a13c67330d2b8645c3ba","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-31","metricId":"Score","value":0.87,"rawValue":"0.87","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=68;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3335bc7af2fcab207b5517ac","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-31","metricId":"Score","value":0.84,"rawValue":"0.84","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=84;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-5ca1c67c930f7f867757094c","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-31","metricId":"Score","value":0.89,"rawValue":"0.89","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=58;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-ac2c47c9331d639d0604a768","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_none","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-31","metricId":"Score","value":0.11833333333333335,"rawValue":"0.11833333333333335","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=217;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-9bac043bc9e6196025ce6c05","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-08-13","metricId":"Score","value":0.8716666666666666,"rawValue":"0.8716666666666666","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=67;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-de6fdc3410d82f3f67eb192f","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-08-13","metricId":"Score","value":0.905,"rawValue":"0.905","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=55;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6fc11ebe4ba54037b5ba7222","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-08-13","metricId":"Score","value":0.9,"rawValue":"0.9","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=56;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-30c2b461686cb6b7bc3c252a","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_none","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-08-13","metricId":"Score","value":0.13,"rawValue":"0.13","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=214;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-60f505378d4d11b23b5bd5fb","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek/deepseek-v3.2","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-01","metricId":"Score","value":0.57,"rawValue":"0.57","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=130;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f087b8424109b96a101d4bca","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro_32K","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-06-17","metricId":"Score","value":0.37,"rawValue":"0.37","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=159;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3b00ed1c1bbdaa32322c8898","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-02-19","metricId":"Score","value":0.98,"rawValue":"0.98","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=4;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-19d1c4ac251ea57ad24c5859","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.535,"rawValue":"0.535","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=136;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-155d234bff6f2459cbd766f0","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.17,"rawValue":"0.17","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=200;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-67ca3110e53a6e351685b3d6","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.3233333333333333,"rawValue":"0.3233333333333333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=170;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ca435d5c44d6fd5387afe41b","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_minimal","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.075,"rawValue":"0.075","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=221;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3924f7470bc98ebbf3f5cbd5","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-05-19","metricId":"Score","value":0.925,"rawValue":"0.925","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=36;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2369d625516e3e6c5ab9a6d7","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_minimal","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-05-19","metricId":"Score","value":0.4883,"rawValue":"0.4883","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=140;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-cf4d5b4adebc0a780ac5c7ca","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.9116666666666666,"rawValue":"0.9116666666666666","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=47;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ff83671bf3978ab9753ecf0d","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.765,"rawValue":"0.765","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=96;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ada0dc3b83ab24d08cb06f2b","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.8316666666666666,"rawValue":"0.8316666666666666","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=85;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-bcd87f75ce0129382e33e78f","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_minimal","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.34500000000000003,"rawValue":"0.34500000000000003","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=161;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-41f98e79f47571a7d8284379","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-08-13","metricId":"Score","value":0.9550000000000001,"rawValue":"0.9550000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=17;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f7783d2bd3c04f2d019ac9c2","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-08-13","metricId":"Score","value":0.8516666666666666,"rawValue":"0.8516666666666666","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=81;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-394d4d958f3d59f60fc605b0","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-08-13","metricId":"Score","value":0.9116666666666666,"rawValue":"0.9116666666666666","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=48;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-45d45b127eca02a346b75b5c","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-02-11","metricId":"Score","value":0.44670000000000004,"rawValue":"0.44670000000000004","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=146;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0daf3857ec28f28193ffa1c9","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_unknown","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-06-16","metricId":"Score","value":0.77,"rawValue":"0.77","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=93;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c1579dd52541904e11fe5826","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.6567000000000001,"rawValue":"0.6567000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=113;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-286da57b80e75b89bf268b96","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.44,"rawValue":"0.44","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=148;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-06391f3c10451c32eb1a6144","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.5617,"rawValue":"0.5617","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=132;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-529313ee5731defdc55952e8","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_minimal","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.06,"rawValue":"0.06","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=223;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ef9efa68e5b721b7a382afce","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.5433,"rawValue":"0.5433","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=134;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-86eb4523081d74fb290b867d","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.2633,"rawValue":"0.2633","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=185;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-e2f30b147db1f9159f1d7901","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.3733000000000001,"rawValue":"0.3733000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=157;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-aea872d4bfdd5bc09ab490c1","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_minimal","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.0533,"rawValue":"0.0533","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=229;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-e4af2814eaf1d70d5c3135aa","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.1667,"rawValue":"0.1667","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=202;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-856a7e09f415f711465aa3d2","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.0404,"rawValue":"0.0404","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=233;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-70fc2dc9ac73808511170124","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.2071,"rawValue":"0.2071","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=197;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-44df386973e61c93c8f39add","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_minimal","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.015,"rawValue":"0.015","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=235;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-7b450a6bb51fd15ea8ed19c7","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-13","metricId":"Score","value":0.728,"rawValue":"0.728","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=105;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-3c32cf6b95a95a22aa30fae7","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-13","metricId":"Score","value":0.7283,"rawValue":"0.7283","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=104;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-fde8716ed0017a652d220491","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-13","metricId":"Score","value":0.33170000000000005,"rawValue":"0.33170000000000005","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=166;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-da2b9b8e2e25a0bb4ac4f398","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-13","metricId":"Score","value":0.577,"rawValue":"0.577","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=126;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-8d5a663597ac61c1816d3378","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-13","metricId":"Score","value":0.5767,"rawValue":"0.5767","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=128;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-b825c512efcc710c65fb6288","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_none","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-11-13","metricId":"Score","value":0.058300000000000005,"rawValue":"0.058300000000000005","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=225;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ef17150dd6e6c133462db03e","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.7867000000000001,"rawValue":"0.7867000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=91;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-53867f3bb07d9e269134fe30","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.5567,"rawValue":"0.5567","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=133;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-38ac69b2b7fc3918e6f63267","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.7267,"rawValue":"0.7267","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=106;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-3470e82e1f9a025b7c6b7ee5","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_none","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.1233,"rawValue":"0.1233","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=215;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-6595b92f4627a41fbb532c91","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.1233,"rawValue":"0.1233","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=216;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-65a0e7e481558817ea99e8ee","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.8617,"rawValue":"0.8617","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=75;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-5d5006115f7d55d131a3e6d1","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.862,"rawValue":"0.862","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=73;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4,"evidenceUrl":"https://arcprize.org/leaderboard"},{"id":"pub-1b6c2754064acac6176f7533","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.8617,"rawValue":"0.8617","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=76;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-ed88efcd97c03760ac750eb5","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.8567,"rawValue":"0.8567","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=79;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7546ee75a473428371fed1c8","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.8117000000000001,"rawValue":"0.8117000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=86;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-a9ae426d15b49c2ce135a4b7","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.905,"rawValue":"0.905","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=51;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-5187c3c58471fe47c7e83ef9","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.905,"rawValue":"0.905","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=53;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-0e4f6c65e0a544b80c69a80e","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.905,"rawValue":"0.905","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=52;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-6110ac71556f31c91dba30c8","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-05","metricId":"Score","value":0.9267000000000001,"rawValue":"0.9267000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=35;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-c77f042f4a970c3e588a1d2e","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-05","metricId":"Score","value":0.6817000000000001,"rawValue":"0.6817000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=110;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-2875c7f66b4f6a5502079dc5","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-05","metricId":"Score","value":0.8617,"rawValue":"0.8617","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=74;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-dd211f4bd5ebd48c170e6e4c","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-05","metricId":"Score","value":0.9367000000000001,"rawValue":"0.9367000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=31;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4274ed88a2f3a3f6209af996","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-05","metricId":"Score","value":0.9367000000000001,"rawValue":"0.9367000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=30;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-1f74b315513513a962121d69","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.58,"rawValue":"0.58","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=125;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-8d324856b83bb12194a98b8e","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.13,"rawValue":"0.13","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=213;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-f714bfde6037c5753410ce06","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.4083,"rawValue":"0.4083","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=152;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c4ed74b566498b852bcf42b0","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.6367,"rawValue":"0.6367","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=119;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-62ccfe44d6ae91e4a3510b31","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.6367,"rawValue":"0.6367","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=118;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-8babfe8f68fa7c1a61a8e74a","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.38170000000000004,"rawValue":"0.38170000000000004","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=156;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-98e70bd32d0eb06dc77d8cb7","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.1833,"rawValue":"0.1833","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=198;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-d2dd040e69a43bbec0f552fa","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.33,"rawValue":"0.33","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=169;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a96ea052472294c6b93002c0","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.515,"rawValue":"0.515","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=139;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-1d5d21f0a580b4bad8da570e","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.515,"rawValue":"0.515","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=138;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-5113a12fcac9e19f04172db5","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-05","metricId":"Score","value":0.9450000000000001,"rawValue":"0.9450000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=22;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-17c26d7e427875280da19040","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-03-05","metricId":"Score","value":0.9450000000000001,"rawValue":"0.9450000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=25;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ee682019c6b77b2f6ba88e2b","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.965,"rawValue":"0.965","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=11;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b6af20368570e23c8676c824","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.9500000000000001,"rawValue":"0.9500000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=20;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-9cd651701168b9ab669a42a1","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.9500000000000001,"rawValue":"0.9500000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=19;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-2171d9a0a0f9c81288a815af","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.9450000000000001,"rawValue":"0.9450000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=23;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-69df6dd9b88ebf4f1d7fb5bd","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.7617,"rawValue":"0.7617","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=97;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-ecf9974a88006220c5b7b947","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.9217000000000001,"rawValue":"0.9217000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=40;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-81701f05a157f227ace52c5c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.9500000000000001,"rawValue":"0.9500000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=18;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-2a7d01499ad6e28cf13d4150","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.9500000000000001,"rawValue":"0.9500000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=21;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-9bd32d0a1a0a2d66fa4039e1","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.765,"rawValue":"0.765","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=95;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-87350f2d457c80a7b1b6a2b2","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.3417,"rawValue":"0.3417","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=163;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-e8fee2f8e9790ecec3a91846","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.88,"rawValue":"0.88","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=60;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c57a0a45658f8a978c05fa7c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.565,"rawValue":"0.565","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=131;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-e1c8d003b3e1adf4a7786e87","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.8767,"rawValue":"0.8767","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=62;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-8f9e36c7b6bc3526ea6a30c6","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.8767,"rawValue":"0.8767","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=61;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-9bc7924b418bdc6f08cd7060","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.97,"rawValue":"0.97","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=10;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-b777498021196563d60a7fda","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.745,"rawValue":"0.745","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=103;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b68412abff564dfc06807d28","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.965,"rawValue":"0.965","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=12;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-f3c49247cbb192bd6c9dc1da","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.925,"rawValue":"0.925","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=38;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-be6d88916841ebf1b5632acf","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.975,"rawValue":"0.975","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=9;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b6e3e9e302588b38a714f7be","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.975,"rawValue":"0.975","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=5;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-a26d55b00518946bff93493a","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.92,"rawValue":"0.92","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=43;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-add26800c1c1397fae47f584","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.6017,"rawValue":"0.6017","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=121;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-d031af764efae46b59f77375","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.965,"rawValue":"0.965","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=13;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3db7f03f115da7901cb8b57c","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.77,"rawValue":"0.77","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=94;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-bf18d823f0ebdbacc42880cc","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.9400000000000001,"rawValue":"0.9400000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=28;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-8fa0345c80983df48472c76a","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.9400000000000001,"rawValue":"0.9400000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=26;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-6459dc5c49cc15fb2245afd8","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2025-09-19","metricId":"Score","value":0.485,"rawValue":"0.485","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=141;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-90c8314b9e28bd9b23fd5bc7","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-08","metricId":"Score","value":0.8567,"rawValue":"0.8567","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=80;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ccdb7228eba11196f916880b","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-08","metricId":"Score","value":0.7917000000000001,"rawValue":"0.7917000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=90;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f8b14a9ff5373a4d0816ba78","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-08","metricId":"Score","value":0.8717,"rawValue":"0.8717","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=66;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1b2db955b1b0c9ac59267b64","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-08-12","metricId":"Score","value":0.87,"rawValue":"0.87","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=70;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-3455892469fe891dbcf6d2bb","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-08-12","metricId":"Score","value":0.7483333333333334,"rawValue":"0.7483333333333334","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=102;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-9d924ea17968c2726d75faec","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_medium","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-08-12","metricId":"Score","value":0.875,"rawValue":"0.875","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=65;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f8af06f7e7cb8dea847500b5","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_xhigh","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-08-12","metricId":"Score","value":0.87,"rawValue":"0.87","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=69;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1babee7ca30fdbd788ef5fe7","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-01-27","metricId":"Score","value":0.6533,"rawValue":"0.6533","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=115;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3c902f27dd8a228df76a3527","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_high","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-16","metricId":"Score","value":0.8666666666666666,"rawValue":"0.8666666666666666","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=71;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-436f0747fb4628655777bad2","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_low","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-16","metricId":"Score","value":0.6566666666666666,"rawValue":"0.6566666666666666","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=114;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-332ab300e5da6fc764e56c7f","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-arc_agi_external","benchmarkVersion":"2026-07-16","metricId":"Score","value":0.9450000000000001,"rawValue":"0.9450000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_external.csv:row=24;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-d719f4be0a797ec4a8216cbf","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-balrog_external","benchmarkVersion":"2025-10-15","metricId":"Average progress","value":31.2,"rawValue":"0.312","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"balrog_external.csv:row=16;column=Average progress","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://balrogai.com/"},{"id":"pub-61f7acd0546142a850ebe33d","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_1K","benchmarkId":"epoch-balrog_external","benchmarkVersion":"2025-10-15","metricId":"Average progress","value":31.2,"rawValue":"0.312","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"balrog_external.csv:row=17;column=Average progress","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://balrogai.com/"},{"id":"pub-fd3e7ced772190f869d6fe92","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-balrog_external","benchmarkVersion":"2025-11-24","metricId":"Average progress","value":43.5,"rawValue":"0.435","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"balrog_external.csv:row=7;column=Average progress","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://balrogai.com/"},{"id":"pub-c233a446c9dcd99a002ca4ab","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_64K","benchmarkId":"epoch-balrog_external","benchmarkVersion":"2025-11-24","metricId":"Average progress","value":43.0,"rawValue":"0.43","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"balrog_external.csv:row=9;column=Average progress","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"evidenceUrl":"https://balrogai.com/"},{"id":"pub-ec7253b14d7248ba527a62fc","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-balrog_external","benchmarkVersion":"2025-11-24","metricId":"Average progress","value":43.0,"rawValue":"0.43","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"balrog_external.csv:row=10;column=Average progress","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://balrogai.com/"},{"id":"pub-abad253da134aaa7f04fa6de","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-balrog_external","benchmarkVersion":"2025-06-17","metricId":"Average progress","value":33.5,"rawValue":"0.335","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"balrog_external.csv:row=12;column=Average progress","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://balrogai.com/"},{"id":"pub-ea058741e5d649f0caa04f2e","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-balrog_external","benchmarkVersion":"2026-02-19","metricId":"Average progress","value":56.99999999999999,"rawValue":"0.57","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"balrog_external.csv:row=3;column=Average progress","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://balrogai.com/"},{"id":"pub-5abba5960ca2cad05d096c07","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-balrog_external","benchmarkVersion":"2026-02-19","metricId":"Average progress","value":56.89999999999999,"rawValue":"0.569","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"balrog_external.csv:row=4;column=Average progress","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://balrogai.com/"},{"id":"pub-d30a30bd8249a2e468b49a3a","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_minimal","benchmarkId":"epoch-balrog_external","benchmarkVersion":"2025-08-07","metricId":"Average progress","value":32.8,"rawValue":"0.32799999999999996","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"balrog_external.csv:row=13;column=Average progress","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://balrogai.com/"},{"id":"pub-758a587ef6d62a870c487bad","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.3861171366594359,"rawValue":"0.3861171366594359","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=2;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d3480cc1435db9a552f8544e","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2025-10-15","metricId":"Score","value":0.0,"rawValue":"0.0","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=15;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-37d74da669f6cb7aacbf0999","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.245119305856833,"rawValue":"0.245119305856833","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=10;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cf772f4e3a44776526a002ae","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.14533622559652917,"rawValue":"0.14533622559652917","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=12;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-092033e01fa8df72bf1b8b26","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.30368763557483736,"rawValue":"0.30368763557483736","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=20;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e17ce06f87f1859f7423f26a","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-02-17","metricId":"Score","value":0.06724511930585665,"rawValue":"0.06724511930585665","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=13;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-07f52eeffb8c60339b595faa","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-06-30","metricId":"Score","value":0.24945770065075923,"rawValue":"0.24945770065075923","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=24;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3b69a56975eb670d9ad158b8","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-02-19","metricId":"Score","value":0.2646420824295011,"rawValue":"0.2646420824295011","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=9;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f4f3e1e8a4963135700f916e","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-05-19","metricId":"Score","value":0.336225596529284,"rawValue":"0.336225596529284","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=4;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-861e4192c89e3e20610adf9b","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.31236442516268986,"rawValue":"0.31236442516268986","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=19;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e40723742a8954f1a33f6760","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-03-05","metricId":"Score","value":0.27114967462039047,"rawValue":"0.27114967462039047","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=8;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f7c44b2d16a956f5e9a593a5","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.36225596529284154,"rawValue":"0.36225596529284154","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=3;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b5040e71b0638941f8ecf4b8","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.22559652928416485,"rawValue":"0.22559652928416485","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=11;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-257fa9d4ac3e30b311b91136","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.336225596529284,"rawValue":"0.336225596529284","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=5;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e552ecf421e02294605dd316","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.30802603036876364,"rawValue":"0.30802603036876364","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=6;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ce9481f0e958bc6e35f3289e","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-07-08","metricId":"Score","value":0.2733188720173536,"rawValue":"0.2733188720173536","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=7;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9bb2b014e4555fa48e666901","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-08-12","metricId":"Score","value":0.33188720173535774,"rawValue":"0.33188720173535774","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=23;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0f1bce64825a25ea615ee8cf","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-04-20","metricId":"Score","value":0.03904555314533626,"rawValue":"0.03904555314533626","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=14;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e4cc889d94545205098adab1","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_unknown","benchmarkId":"epoch-blueprint_bench_2_external","benchmarkVersion":"2026-07-16","metricId":"Score","value":0.29501084598698485,"rawValue":"0.29501084598698485","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=21;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-55ebce056f60240ad7395b1a","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-btf3_external","benchmarkVersion":"2026-06-09","metricId":"Pooled score","value":0.13,"rawValue":"0.13","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"btf3_external.csv:row=3;column=Pooled score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"FutureSearch agent","subject_type":"system"},"harnessId":"FutureSearch agent","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e8341041bed00e2e16a36d2f","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_high","benchmarkId":"epoch-btf3_external","benchmarkVersion":"2026-05-28","metricId":"Pooled score","value":0.14,"rawValue":"0.14","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"btf3_external.csv:row=7;column=Pooled score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"FutureSearch agent","subject_type":"system"},"harnessId":"FutureSearch agent","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-38652d412e7a3f501848776a","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_high","benchmarkId":"epoch-btf3_external","benchmarkVersion":"2026-05-28","metricId":"Pooled score","value":0.137,"rawValue":"0.137","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"btf3_external.csv:row=6;column=Pooled score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Vendor agent SDK","subject_type":"system"},"harnessId":"Vendor agent SDK","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-82f64451d1ed293a97b18232","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_xhigh","benchmarkId":"epoch-btf3_external","benchmarkVersion":"2026-05-28","metricId":"Pooled score","value":0.13,"rawValue":"0.13","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"btf3_external.csv:row=2;column=Pooled score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"FutureSearch agent","subject_type":"system"},"harnessId":"FutureSearch agent","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-aa726b5890057494f312b12a","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_xhigh","benchmarkId":"epoch-btf3_external","benchmarkVersion":"2026-07-24","metricId":"Pooled score","value":0.118,"rawValue":"0.118","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"btf3_external.csv:row=10;column=Pooled score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"FutureSearch agent","subject_type":"system"},"harnessId":"FutureSearch agent","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d523de5cfbc7a718eecba119","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_xhigh","benchmarkId":"epoch-btf3_external","benchmarkVersion":"2026-06-30","metricId":"Pooled score","value":0.154,"rawValue":"0.154","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"btf3_external.csv:row=9;column=Pooled score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"FutureSearch agent","subject_type":"system"},"harnessId":"FutureSearch agent","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f306077ce0ea72e00e8f41fb","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-btf3_external","benchmarkVersion":"2026-04-23","metricId":"Pooled score","value":0.143,"rawValue":"0.143","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"btf3_external.csv:row=8;column=Pooled score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"FutureSearch agent","subject_type":"system"},"harnessId":"FutureSearch agent","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bd690d0d28ad1ad42e00e672","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-btf3_external","benchmarkVersion":"2026-04-23","metricId":"Pooled score","value":0.134,"rawValue":"0.134","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"btf3_external.csv:row=4;column=Pooled score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Vendor agent SDK","subject_type":"system"},"harnessId":"Vendor agent SDK","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-bdf8b7aa1fcf5ecf01b400c4","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_high","benchmarkId":"epoch-btf3_external","benchmarkVersion":"2026-07-09","metricId":"Pooled score","value":0.135,"rawValue":"0.135","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"btf3_external.csv:row=5;column=Pooled score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"FutureSearch agent","subject_type":"system"},"harnessId":"FutureSearch agent","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6c4da5d1e5f3a69a5dd5320d","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-btf3_external","benchmarkVersion":"2026-07-09","metricId":"Pooled score","value":0.137,"rawValue":"0.137","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"btf3_external.csv:row=11;column=Pooled score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"FutureSearch agent","subject_type":"system"},"harnessId":"FutureSearch agent","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-038ef47437cf98c11b99748d","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-06-01","metricId":"mean_score","value":14.000000000000002,"rawValue":"0.14","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=8;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-012fc5e65fcf647dcc8833a0","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-06-09","metricId":"mean_score","value":41.0,"rawValue":"0.41","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=81;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-d94276a89b90000e0cd19a83","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-06-09","metricId":"mean_score","value":28.000000000000004,"rawValue":"0.28","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=80;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-fe5bec1d97eabe441c39785a","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-06-09","metricId":"mean_score","value":41.0,"rawValue":"0.41","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=127;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-273d4414f198e3a071cfc5d3","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_32K","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-10-15","metricId":"mean_score","value":8.0,"rawValue":"0.08","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=91;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3de9ea73e57e6634d9d46f1b","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-11-24","metricId":"mean_score","value":4.0,"rawValue":"0.04","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=79;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-a2d3526f9b71fed1f32ad3df","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_32K","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-11-24","metricId":"mean_score","value":12.0,"rawValue":"0.12","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=162;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d7917ba7612edfc4a880fe5f","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_120K","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-05","metricId":"mean_score","value":13.0,"rawValue":"0.13","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=138;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-26746e3d6dab1441ab9f57db","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_32K","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-05","metricId":"mean_score","value":17.0,"rawValue":"0.17","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=143;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-9bbd202b92ec72f3e04b2004","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_64K","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-05","metricId":"mean_score","value":10.0,"rawValue":"0.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=142;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-e43e2a0c0db2dc13bad94eac","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-05","metricId":"mean_score","value":14.000000000000002,"rawValue":"0.14","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=62;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-56182c935b3ff9a4c85fb1a8","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-16","metricId":"mean_score","value":20.0,"rawValue":"0.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=109;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-619d4e18066ee2c0b0e2bae4","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-16","metricId":"mean_score","value":7.000000000000001,"rawValue":"0.07","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=78;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1b8634a19537dd471df5460e","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_xhigh","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-16","metricId":"mean_score","value":30.0,"rawValue":"0.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=133;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-d6398eb60d1912b520a4e077","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-05-28","metricId":"mean_score","value":28.999999999999996,"rawValue":"0.29","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=77;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-de6a1a90c76bf96a969988b8","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-05-28","metricId":"mean_score","value":34.0,"rawValue":"0.34","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=128;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8370278f939efc16521e6198","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-05-28","metricId":"mean_score","value":13.0,"rawValue":"0.13","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=76;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-62d94ddb70e817ab22c38604","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-24","metricId":"mean_score","value":33.0,"rawValue":"0.33","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=75;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3324addc998ad9eb86234dbe","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-24","metricId":"mean_score","value":20.0,"rawValue":"0.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=74;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-4fe5acb3f3f99e836fb3e592","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-24","metricId":"mean_score","value":42.0,"rawValue":"0.42","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=88;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-a95fe7cdbcc3e3e90240d570","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-09-29","metricId":"mean_score","value":4.0,"rawValue":"0.04","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=72;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a2a8ce9e61a1639dca94f3bd","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-09-29","metricId":"mean_score","value":12.0,"rawValue":"0.12","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=157;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e0b1d5b80f8c5a682687b0da","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_32K","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-17","metricId":"mean_score","value":13.0,"rawValue":"0.13","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=139;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7938d65003a5c6220a7ddb85","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-17","metricId":"mean_score","value":5.0,"rawValue":"0.05","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=111;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-0a9f27849af087e93a8536b3","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-17","metricId":"mean_score","value":3.0,"rawValue":"0.03","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=73;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-1856ce8e0c30e4a3b6474808","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_medium","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-17","metricId":"mean_score","value":8.0,"rawValue":"0.08","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=114;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-3875977ae2a6796636145ee2","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-06-30","metricId":"mean_score","value":16.0,"rawValue":"0.16","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=123;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d5ebdafca01b76b9060f7fdf","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_xhigh","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-06-30","metricId":"mean_score","value":35.0,"rawValue":"0.35","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=122;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e472029872c07b6d676b48d1","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-chat","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-12-01","metricId":"mean_score","value":1.0,"rawValue":"0.01","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=92;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-51a404e9c6fb7e191c5dd8ed","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-31","metricId":"mean_score","value":33.0,"rawValue":"0.33","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=86;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-739377b74349e3cf9260585f","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-08-13","metricId":"mean_score","value":47.0,"rawValue":"0.47","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=3;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e4c8afb6e4203b5df65703bb","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-24","metricId":"mean_score","value":13.0,"rawValue":"0.13","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=61;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c6c2c894f5a8077059360909","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-24","metricId":"mean_score","value":20.0,"rawValue":"0.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=126;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-160b76a8645192d7a11582b5","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-24","metricId":"mean_score","value":7.000000000000001,"rawValue":"0.07","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=60;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-61f944c64d98c7fca0c00f6c","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"fireworks/kimi-k2p5","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-01-27","metricId":"mean_score","value":12.0,"rawValue":"0.12","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=146;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5fba340f8eded73510d0db12","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-06-17","metricId":"mean_score","value":20.0,"rawValue":"0.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=163;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a959816d0bd34bd235e4caa0","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-19","metricId":"mean_score","value":55.00000000000001,"rawValue":"0.55","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=140;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1fd973a83793aeac96fa04d4","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-19","metricId":"mean_score","value":49.0,"rawValue":"0.49","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=67;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b06b5017adbbe611069117eb","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":22.0,"rawValue":"0.22","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=65;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-878288b19809c4055a71ca20","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":18.0,"rawValue":"0.18","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=64;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-3493f8118b9e5b1be3dba352","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_minimal","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":21.0,"rawValue":"0.21","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=63;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-8b855b61acf0f9b564a8d3cd","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-05-19","metricId":"mean_score","value":50.0,"rawValue":"0.5","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=129;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c1a03bbdb8c9e3f08de0a562","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-05-19","metricId":"mean_score","value":45.0,"rawValue":"0.45","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=66;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-73d9bf60a3dcce51423d7784","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_minimal","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-05-19","metricId":"mean_score","value":43.0,"rawValue":"0.43","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=107;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-84d31b64fe0db4172fd4f2e0","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":40.0,"rawValue":"0.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=87;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-4e4883bf4d141da38abcbc05","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":43.0,"rawValue":"0.43","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=42;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-e7897a543b08669324af4218","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_minimal","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":35.0,"rawValue":"0.35","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=41;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e669176b91a1b36b0c1b2d4c","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-08-13","metricId":"mean_score","value":47.0,"rawValue":"0.47","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=4;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e5d0c28a0a0dfe83c5dffc35","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-12-22","metricId":"mean_score","value":6.0,"rawValue":"0.06","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=145;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6a6c698ae833a36eecbc0405","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-11","metricId":"mean_score","value":10.0,"rawValue":"0.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=141;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7a844bfa7f52e8c99bf0127a","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-07","metricId":"mean_score","value":19.0,"rawValue":"0.19","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=10;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2f55ee0845fcfcb446267b0f","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-06-16","metricId":"mean_score","value":21.0,"rawValue":"0.21","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=124;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-f512d9b2b10055c203b126de","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-06-16","metricId":"mean_score","value":6.0,"rawValue":"0.06","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=9;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-007a624638b53d08afbaca5e","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-08-14","metricId":"mean_score","value":21.0,"rawValue":"0.21","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=2;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e2d44feee76e8b9e24907601","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":37.0,"rawValue":"0.37","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=160;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-62804e7c1d1ffb2c4f449890","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":24.0,"rawValue":"0.24","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=105;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-015226ef900922dfaf06cf31","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":28.999999999999996,"rawValue":"0.29","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=104;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-b40cba42260f7fe3e1e45c3f","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_minimal","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":16.0,"rawValue":"0.16","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=102;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-9f1bfa0185d1ffee7a54861f","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":30.0,"rawValue":"0.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=50;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-754007ceb7c519bb9a03e7a7","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":12.0,"rawValue":"0.12","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=51;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-cef9b7f635c61a8aba82a82d","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_minimal","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":7.000000000000001,"rawValue":"0.07","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=49;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b9b0cd80a89bf356c10a0689","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":27.0,"rawValue":"0.27","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=55;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4a3fdf2f940cd82005c837a5","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":15.0,"rawValue":"0.15","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=110;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-2b7757f20ed9be4863eb7dfe","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_minimal","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":1.0,"rawValue":"0.01","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=112;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-9019ff7263d99e405880fe2a","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-11-13","metricId":"mean_score","value":32.0,"rawValue":"0.32","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=159;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-22f4ad1638bb97f0cafa9231","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-11-13","metricId":"mean_score","value":14.000000000000002,"rawValue":"0.14","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=39;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-a11b58d8218e5f71dc0b8aac","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-11-13","metricId":"mean_score","value":17.0,"rawValue":"0.17","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=27;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3dad1cb2980de36c93aeec9c","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-12-11","metricId":"mean_score","value":40.0,"rawValue":"0.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=152;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-35dc567736ce613ea58ea4bb","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-12-11","metricId":"mean_score","value":23.0,"rawValue":"0.23","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=153;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-292f5cd8d48cdc8066c0e079","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-12-11","metricId":"mean_score","value":40.0,"rawValue":"0.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=151;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-972ae6328adf3a105f9b867f","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-12-11","metricId":"mean_score","value":4.0,"rawValue":"0.04","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=113;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-156eb016f1a1f8ba835ac7ec","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2025-12-11","metricId":"mean_score","value":49.0,"rawValue":"0.49","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=149;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-6e41bc566f8ec9c55a2169bb","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":38.0,"rawValue":"0.38","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=100;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-69fa49bc70d9e933bba8e992","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":20.0,"rawValue":"0.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=101;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-d9eb62ee83004c9fcce37870","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_medium","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":38.0,"rawValue":"0.38","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=98;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d38a947c1001526c013812b7","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":5.0,"rawValue":"0.05","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=99;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b1fb2b8c2a28a5803853ea48","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":44.0,"rawValue":"0.44","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=137;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-52dc8cee916905a208fff0b0","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":18.0,"rawValue":"0.18","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=134;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ba0ccd40194a00a97226c7f4","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":3.0,"rawValue":"0.03","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=21;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0dbd64c73b69e95c87fbea37","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_xhigh","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":24.0,"rawValue":"0.24","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=14;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-b86f789a60d1381c3faa3bb7","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":30.0,"rawValue":"0.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=135;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-6ea57b7bf868e516445b26f5","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":17.0,"rawValue":"0.17","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=58;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-df013abc0b830be0fc1f5203","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":3.0,"rawValue":"0.03","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=56;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e2ab15ce31112bcfc8aef5e6","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_xhigh","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":58.599999999999994,"rawValue":"0.586","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=136;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6c80d906448ca7586a661cf8","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-05-05","metricId":"mean_score","value":12.0,"rawValue":"0.12","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=85;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-864d7fb20c43b3a89fc5c26b","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-23","metricId":"mean_score","value":26.0,"rawValue":"0.26","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=24;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-035053f23e048361bb0d93a9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-23","metricId":"mean_score","value":10.0,"rawValue":"0.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=16;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-9b178a251ed2864b759fd163","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":21.0,"rawValue":"0.21","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=46;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-e52d4eace77f6469c53d6033","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":40.0,"rawValue":"0.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=119;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-84280ae74f22bfaab29dbade","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":2.0,"rawValue":"0.02","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=31;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-1216d66f44090ca4ddfa880c","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":27.0,"rawValue":"0.27","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=22;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-9f59356d9396c0738bfdfa00","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":55.00000000000001,"rawValue":"0.55","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=120;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-426ea876e1d0b1b4de802cf3","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":7.000000000000001,"rawValue":"0.07","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=15;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-617e1bb82062f6c832c5e36e","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_promax","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":64.0,"rawValue":"0.64","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=117;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-4eedcf3959632f234bed6ecf","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":22.0,"rawValue":"0.22","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=29;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-f5131329ba6c1f34bd5fe140","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":54.0,"rawValue":"0.54","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=118;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6217ece18b4de3c7abf4ded9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":5.0,"rawValue":"0.05","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=20;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e35565da6b8b2cc9a7d42be4","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-08","metricId":"mean_score","value":36.0,"rawValue":"0.36","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=121;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-97b8f011b3fe97e8cfaa789b","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-08-12","metricId":"mean_score","value":40.0,"rawValue":"0.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=7;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c529cd6b85a779c433971820","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_xhigh","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-08-12","metricId":"mean_score","value":31.0,"rawValue":"0.31","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=6;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-285dc1507c9181c1da720657","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-20","metricId":"mean_score","value":26.0,"rawValue":"0.26","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=130;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d44ea23167d6bce5bf85290e","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-06-12","metricId":"mean_score","value":21.0,"rawValue":"0.21","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=28;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-804c032cc29e3ae56b72ecc4","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_high","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-16","metricId":"mean_score","value":25.0,"rawValue":"0.25","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=18;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-20cf1f3bf60009f1656e670e","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_low","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-16","metricId":"mean_score","value":20.0,"rawValue":"0.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=17;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-d377ae85b517108c0f05247c","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-16","metricId":"mean_score","value":39.0,"rawValue":"0.39","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=90;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3061e08f3c4600f6aef9bb44","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-13","metricId":"mean_score","value":12.0,"rawValue":"0.12","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=37;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1cfad99a570bd1845d546dec","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-13","metricId":"mean_score","value":13.0,"rawValue":"0.13","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=38;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-1037ba670fe4ed7d901f217f","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-25","metricId":"mean_score","value":21.0,"rawValue":"0.21","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=57;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2758380a83b6389098371950","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-02-16","metricId":"mean_score","value":22.0,"rawValue":"0.22","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=40;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6e8f764d2a34a2086f49bd95","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-22","metricId":"mean_score","value":22.0,"rawValue":"0.22","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=35;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-eca80e11a36d86de141c31bb","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-22","metricId":"mean_score","value":9.0,"rawValue":"0.09","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=36;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5aaca41a32519cfbafdfba8f","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-14","metricId":"mean_score","value":26.0,"rawValue":"0.26","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=43;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-73b82998ceb9d5374d2f507a","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-14","metricId":"mean_score","value":4.0,"rawValue":"0.04","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=44;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-987038b6d3129780a60d1a63","canonicalModelId":"qwen/qwen3.6-flash@2026-04-27","modelName":"Qwen3.6 Flash","modelRef":"qwen3.6-flash","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-27","metricId":"mean_score","value":20.0,"rawValue":"0.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=45;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2d1c5e4bfac2957ba947d664","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-04-20","metricId":"mean_score","value":20.0,"rawValue":"0.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=23;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0212f15a7ade3ccb1fb828c9","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-03-31","metricId":"mean_score","value":17.0,"rawValue":"0.17","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=32;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a774e3fcab957c61cc7d59e5","canonicalModelId":"qwen/qwen3.7-flash@2026-07-27","modelName":"Qwen3.7 Flash","modelRef":"qwen3.7-flash","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-27","metricId":"mean_score","value":23.0,"rawValue":"0.23","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=52;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9060063e875f3d4e23bfe72b","canonicalModelId":"qwen/qwen3.7-flash@2026-07-27","modelName":"Qwen3.7 Flash","modelRef":"qwen3.7-flash_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-07-27","metricId":"mean_score","value":9.0,"rawValue":"0.09","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=53;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-7fb8f467104f24ee5dee2a91","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-05-19","metricId":"mean_score","value":19.0,"rawValue":"0.19","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=25;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4d254c603c87a2ecf32ebe74","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-06-02","metricId":"mean_score","value":24.0,"rawValue":"0.24","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=47;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3be60c1cdab6534d2e7e571e","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus_none","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-06-02","metricId":"mean_score","value":9.0,"rawValue":"0.09","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=48;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-2230dab74d046a7fe2505883","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max_xhigh","benchmarkId":"epoch-chess_puzzles","benchmarkVersion":"2026-08-02","metricId":"mean_score","value":28.999999999999996,"rawValue":"0.29","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=84;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-10594b9a39db5759f09049fa","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp_thinking","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2025-09-29","metricId":"Overall","value":0.132,"rawValue":"0.132","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=21;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-51d2513503aada81fa1c8e3c","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2026-02-12","metricId":"Overall","value":0.114,"rawValue":"0.114","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=24;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-85aad312b561e554187e6711","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2025-11-24","metricId":"Overall","value":0.211,"rawValue":"0.211","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=6;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-99d5eed13b0c6f58961db61b","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2026-02-05","metricId":"Overall","value":0.207,"rawValue":"0.207","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=8;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4283888d4a3869944fce6114","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek/deepseek-v3.2","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2025-12-01","metricId":"Overall","value":0.12400000000000001,"rawValue":"0.12400000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=22;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ebe35acf5ebd2b8a479fd1b6","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2026-02-19","metricId":"Overall","value":0.20800000000000002,"rawValue":"0.20800000000000002","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=7;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6e208c8e36b99e5d14d4a4fb","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2026-02-11","metricId":"Overall","value":0.187,"rawValue":"0.187","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=12;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-61cb9f5f4a5a8a35452eedb9","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2025-11-13","metricId":"Overall","value":0.237,"rawValue":"0.237","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=3;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-209c7b0972292e70b64258be","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2025-11-13","metricId":"Overall","value":0.211,"rawValue":"0.211","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=5;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-20e3cdb623a551c7647f1b57","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2025-12-11","metricId":"Overall","value":0.181,"rawValue":"0.181","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=14;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-df5fbd93b93d62878ae9b0e1","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2025-12-11","metricId":"Overall","value":0.182,"rawValue":"0.182","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=13;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-81cba128cc6553d868cfb5e4","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2026-03-05","metricId":"Overall","value":0.279,"rawValue":"0.279","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=2;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fa57392fba8d817011ef883c","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2026-01-27","metricId":"Overall","value":0.193,"rawValue":"0.193","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=11;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dd162e39208984607a217d0d","canonicalModelId":"xiaomi/mimo-v2-pro@current","modelName":"MiMo V2 Pro","modelRef":"mimo-v2-pro","benchmarkId":"epoch-cl_bench_external","metricId":"Overall","value":0.157,"rawValue":"0.157","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=19;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-709cfcae239b6533650a426e","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2026-02-16","metricId":"Overall","value":0.198,"rawValue":"0.198","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=10;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a7ee9b599328570e76cf22cf","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2026-03-31","metricId":"Overall","value":0.203,"rawValue":"0.203","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=9;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-741f8fea80bad2a285278108","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/glm-4.7","benchmarkId":"epoch-cl_bench_external","benchmarkVersion":"2025-12-22","metricId":"Overall","value":0.159,"rawValue":"0.159","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=17;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b2343ed6eaa22c03cb18e038","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp_thinking","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2025-09-29","metricId":"Overall","value":0.095,"rawValue":"0.095","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=15;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c6d90dc1cc0401bceb0125ee","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2026-02-12","metricId":"Overall","value":0.063,"rawValue":"0.063","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=18;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e5521719c8b529123e78d2c1","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_high","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2026-02-05","metricId":"Overall","value":0.17,"rawValue":"0.17","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=6;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-89d8a40bb5c0095db15ab2b4","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2026-02-05","metricId":"Overall","value":0.136,"rawValue":"0.136","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=9;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e97176c852349a176f044238","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_high","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2026-04-24","metricId":"Overall","value":0.135,"rawValue":"0.135","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=10;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1c46707e4d1c97cc6eba0741","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek/deepseek-v3.2","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2025-12-01","metricId":"Overall","value":0.074,"rawValue":"0.074","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=16;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4cb67d2bf8f3cc0b58f10cb7","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2026-02-19","metricId":"Overall","value":0.169,"rawValue":"0.169","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=7;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-40af1f564b92030fc5417c9a","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2025-11-13","metricId":"Overall","value":0.173,"rawValue":"0.173","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=5;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9b1df0fcc3cb0f4fbfe7e2d5","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2026-03-05","metricId":"Overall","value":0.193,"rawValue":"0.193","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=4;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-1a334fa316da81d1837018f3","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_unknown","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2026-03-05","metricId":"Overall","value":0.138,"rawValue":"0.138","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=8;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-eb58b6ed4d3228237378207f","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2026-03-05","metricId":"Overall","value":0.217,"rawValue":"0.217","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=3;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4f3dc4ef2c63847cbd35a3e9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2026-04-23","metricId":"Overall","value":0.222,"rawValue":"0.222","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=2;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-13933aa7ce98a97bb405328a","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2026-01-27","metricId":"Overall","value":0.132,"rawValue":"0.132","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=11;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-af824ce917d90ba80917985f","canonicalModelId":"xiaomi/mimo-v2-pro@current","modelName":"MiMo V2 Pro","modelRef":"mimo-v2-pro","benchmarkId":"epoch-cl_bench_life_external","metricId":"Overall","value":0.069,"rawValue":"0.069","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=17;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fe34fef512bdb793b47a20dc","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2026-02-16","metricId":"Overall","value":0.12400000000000001,"rawValue":"0.12400000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=12;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-be9b21ad1adcf8031248dcf6","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/glm-4.7","benchmarkId":"epoch-cl_bench_life_external","benchmarkVersion":"2025-12-22","metricId":"Overall","value":0.109,"rawValue":"0.109","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_life_external.csv:row=14;column=Overall","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1159f02ccee2cb983abadb75","canonicalModelId":"deepseek/v3.1-terminus@2025-09-22","modelName":"DeepSeek V3.1 Terminus","modelRef":"DeepSeek-V3.1-Terminus","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-09-22","metricId":"Accuracy","value":1.71428571428571,"rawValue":"0.0171428571428571","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=77;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4aba9b883c3e45e9e3a0f740","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp_thinking","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy","value":2.85714285714286,"rawValue":"0.0285714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=72;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8be536e4a29886049a0a4d32","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMax-M2.7","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-03-18","metricId":"Accuracy","value":0.571428571428571,"rawValue":"0.00571428571428571","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=93;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d9148f15daf214e3a3c72a0a","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-06-01","metricId":"Accuracy","value":3.7142857142857104,"rawValue":"0.0371428571428571","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=61;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-063d5043bac8039a73068c18","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-06-09","metricId":"Accuracy","value":28.571428570000002,"rawValue":"0.2857142857","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=8;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d2874b36bed485426b007fd7","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=109;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dd5a88f894b921bfeb495e68","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_unknown","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-05-22","metricId":"Accuracy","value":0.3,"rawValue":"0.003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=94;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-47df187afed1f4b5c5937b20","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-16","metricId":"Accuracy","value":12.0,"rawValue":"0.12","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=39;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f6c5c2e16ddd244d9c7b541e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-05-28","metricId":"Accuracy","value":20.8571428571429,"rawValue":"0.20857142857142902","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=22;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0054761eaeb01936f3a26fc5","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-24","metricId":"Accuracy","value":28.2857142857143,"rawValue":"0.282857142857143","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=9;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-2bf5c7588fa74039e56045e1","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_low","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-24","metricId":"Accuracy","value":23.1428571428571,"rawValue":"0.231428571428571","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=19;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-d807f78b105712d00b6583bb","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-24","metricId":"Accuracy","value":29.142857100000004,"rawValue":"0.291428571","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=6;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0e84bc6035f8d348df1ff5aa","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_medium","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-24","metricId":"Accuracy","value":26.8571428571429,"rawValue":"0.268571428571429","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=13;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-340dcfa87c2c0c548cb4c43a","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_xhigh","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-24","metricId":"Accuracy","value":27.7142857142857,"rawValue":"0.277142857142857","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=10;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-35a8f8ddc2504d8f8091ed66","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy","value":1.1428571428571401,"rawValue":"0.0114285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=85;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3ecec90d521a66702eefc8bf","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-02-17","metricId":"Accuracy","value":3.14285714285714,"rawValue":"0.0314285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=66;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fb97fb4cd24f80b10aaf695d","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-06-30","metricId":"Accuracy","value":1.1428571428571401,"rawValue":"0.0114285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=87;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8d63d36229d4f8024861042c","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-06-30","metricId":"Accuracy","value":16.8571428571429,"rawValue":"0.168571428571429","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=29;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-f4746213142916fc895a3c43","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-31","metricId":"Accuracy","value":16.5714285714286,"rawValue":"0.165714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=31;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-401f0284df6df2ba0232578b","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-24","metricId":"Accuracy","value":3.42857142857143,"rawValue":"0.0342857142857143","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=64;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-c78d720c462c092d80cdb3fc","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-24","metricId":"Accuracy","value":7.14285714285714,"rawValue":"0.0714285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=53;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-c02d64a6a7af840f9c898428","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-24","metricId":"Accuracy","value":10.0,"rawValue":"0.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=44;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d0952d3c1319e9ba862ab5c0","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-24","metricId":"Accuracy","value":12.8571428571429,"rawValue":"0.128571428571429","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=37;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e4823853f00e4ef94403dece","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy","value":1.1,"rawValue":"0.011000000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=89;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e30867950b38a6c52a16a374","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy","value":2.0,"rawValue":"0.02","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=75;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-590c550c9beb7b7f25e3ecb2","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-02-19","metricId":"Accuracy","value":17.714285714285698,"rawValue":"0.177142857142857","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=27;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-20c0c0ba7d41105c95b94495","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-21","metricId":"Accuracy","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=121;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1874a0da2dcd438e40e6b712","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-05-19","metricId":"Accuracy","value":13.1428571428571,"rawValue":"0.131428571428571","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=36;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0981a795aefd43cb32250a51","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-21","metricId":"Accuracy","value":10.5714285714286,"rawValue":"0.105714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=41;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-344259f8d227755512ef1b25","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-12-22","metricId":"Accuracy","value":1.7102040816326498,"rawValue":"0.0171020408163265","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=78;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1ec86b0af8cb9cefd8643761","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-07","metricId":"Accuracy","value":4.57142857142857,"rawValue":"0.0457142857142857","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=59;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b5739bd28f013c6e98d3d396","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-06-16","metricId":"Accuracy","value":20.8571428571429,"rawValue":"0.20857142857142902","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=23;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7908b2fb551fc5b6f86e05b2","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_none","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-06-16","metricId":"Accuracy","value":3.14285714285714,"rawValue":"0.0314285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=70;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-eb07964801bbc4a482e7358d","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":12.6,"rawValue":"0.126","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=38;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-fdc5a8e80a380222e8fbbbf7","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_minimal","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=105;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-98ed20419d96e716ae818177","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=135;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9704985a8223fa9c5fa728c5","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2025-11-13","metricId":"Accuracy","value":4.857142857142859,"rawValue":"0.0485714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=57;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-32a4817b08f9ce71ca1713fd","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-03-05","metricId":"Accuracy","value":23.42857143,"rawValue":"0.23428571430000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=17;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-95d4b802736dd376313e7b35","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_xhigh","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-03-17","metricId":"Accuracy","value":10.0,"rawValue":"0.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=42;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-298f0598adde999db98a9559","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_xhigh","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-03-17","metricId":"Accuracy","value":9.253061224489791,"rawValue":"0.0925306122448979","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=46;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-aa601a4193fd60fbc5e0d452","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_xhigh","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-03-05","metricId":"Accuracy","value":30.0,"rawValue":"0.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=4;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b102a9172041ce4ead9251a1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-05-05","metricId":"Accuracy","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=138;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-f1fe7d3b56a8ef1a5a0a14e4","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_xhigh","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy","value":30.571428571428598,"rawValue":"0.305714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=3;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-224ba0d9f79b652994d162ba","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy","value":25.428571428571402,"rawValue":"0.254285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=16;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-1cede2452e77b1c2d636cf83","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_low","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy","value":8.0,"rawValue":"0.08","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=50;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-22b28089576931d76aca46a1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_medium","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy","value":18.5714285714286,"rawValue":"0.18571428571428603","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=26;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-0784d2bb74ff456da260d1fd","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_none","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy","value":1.42857142857143,"rawValue":"0.0142857142857143","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=80;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-d7f39fccf6b77fd8d12401e6","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy","value":27.142857142857103,"rawValue":"0.271428571428571","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=11;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7249793bfa12922b577941c6","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":16.5714285714286,"rawValue":"0.165714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=30;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-9818494cdb9b8e3345311227","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_low","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":2.57142857142857,"rawValue":"0.0257142857142857","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=73;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-e1c9fd584727cddc1b6c0128","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":20.599999999999998,"rawValue":"0.206","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=24;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-21ca7816cb14ea941843c49c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_medium","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":4.857142857142859,"rawValue":"0.0485714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=58;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-1b453eb2d39d9345b4f182f4","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_none","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":0.28571428571428603,"rawValue":"0.00285714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=103;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-5fa9bb581e07af45639878d5","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_xhigh","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":20.571428571428598,"rawValue":"0.205714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=25;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-c316650f84c3b544c9a3b8f8","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":25.7142857142857,"rawValue":"0.257142857142857","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=15;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-627ec9cbfe866c34ff1f7215","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_low","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":14.857142857142899,"rawValue":"0.148571428571429","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=34;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-ccc970092bcde6cf9d7fa1f9","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":32.300000000000004,"rawValue":"0.323","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=2;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-d2625fce1c45e65ef5a762d6","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_medium","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":22.8571428571429,"rawValue":"0.228571428571429","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=21;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-fae19dbb41a98d31057b40b3","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_none","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":5.14285714285714,"rawValue":"0.0514285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=56;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-50acae49aa956931819cfa05","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":28.599999999999998,"rawValue":"0.286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=7;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-ba982c12d7ac5fcebe8ad13f","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":22.8571428571429,"rawValue":"0.228571428571429","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=20;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-dc4c95816752098656f7f36d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_low","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":9.42857142857143,"rawValue":"0.0942857142857143","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=45;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-337fe229bd3904e2636c7273","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":30.0,"rawValue":"0.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=5;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-0609fc2481cfd284852c34e0","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_medium","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":17.4285714285714,"rawValue":"0.174285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=28;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-9cf01d6ce3b09456a5405257","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_none","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":2.0,"rawValue":"0.02","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=76;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-2e880381f3e286c3410618f5","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":27.142857142857103,"rawValue":"0.271428571428571","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=12;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-c19b0aaf3ec9e6c3edf4459f","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-08","metricId":"Accuracy","value":15.4285714285714,"rawValue":"0.154285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=32;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b060eab6ae1b9cb74ea11da4","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-01-27","metricId":"Accuracy","value":3.14285714285714,"rawValue":"0.0314285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=68;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a5d1cf6b8f1acc16d09c0664","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-20","metricId":"Accuracy","value":8.0,"rawValue":"0.08","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=51;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-86f60ee6f3798adedc268930","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-06-12","metricId":"Accuracy","value":10.0,"rawValue":"0.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=43;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3fecaeae3e7d329b6ab7fb65","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_low","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-16","metricId":"Accuracy","value":3.14285714285714,"rawValue":"0.0314285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=69;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d5c82478bcb864507003f958","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-07-16","metricId":"Accuracy","value":23.400000000000002,"rawValue":"0.234","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=18;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4aa98df35deced777d7849a3","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash","benchmarkId":"epoch-critpt_external","metricId":"Accuracy","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=111;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-77a3d7905fcb886b7a616d8f","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"epoch-critpt_external","metricId":"Accuracy","value":3.7142857142857104,"rawValue":"0.0371428571428571","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=63;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8c4561c09ef66404337c01ce","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"epoch-critpt_external","metricId":"Accuracy","value":4.0,"rawValue":"0.04","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=60;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e03fe337cf86b3bebcc36387","canonicalModelId":"qwen/qwen3-coder-next@2026-02-03","modelName":"Qwen3-Coder-Next","modelRef":"qwen3-coder-next","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-02-02","metricId":"Accuracy","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=130;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-10c7a6eaa5632b13ec94966a","canonicalModelId":"qwen/qwen3.5-9b@current","modelName":"Qwen3.5 9B","modelRef":"qwen3.5-9B","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-02-24","metricId":"Accuracy","value":0.28571428571428603,"rawValue":"0.00285714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=95;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-55540e32dfd6ea8ea272d071","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-04-14","metricId":"Accuracy","value":0.28571428571428603,"rawValue":"0.00285714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=96;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-99a8ead606b2a73ac59d546a","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-03-31","metricId":"Accuracy","value":2.85714285714286,"rawValue":"0.0285714285714286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=71;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9d41c4e5a64e8947e021dd82","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-05-19","metricId":"Accuracy","value":13.4285714285714,"rawValue":"0.134285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=35;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ad89304a1574736ab354c403","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"epoch-critpt_external","benchmarkVersion":"2026-06-02","metricId":"Accuracy","value":9.14285714285714,"rawValue":"0.0914285714285714","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=47;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5ce3da02e40d2dee8ea5d936","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.665,"rawValue":"0.665","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=4;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-20724793eeccb3a367f69b1a","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.621,"rawValue":"0.621","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=42;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-3b30ff2298312767a565c0fb","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.705,"rawValue":"0.705","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=2;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-6071a317f09a227daddd8518","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.652,"rawValue":"0.652","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=36;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ee08db86c10c916b3f989cc7","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_xhigh","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.684,"rawValue":"0.684","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=3;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-df8f4478b9eb1c07fd30939d","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.594,"rawValue":"0.594","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=11;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b21469e21158801c245f7a53","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.483,"rawValue":"0.483","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=23;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-cbe17d566ad9d0e2e6247e2e","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.648,"rawValue":"0.648","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=5;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-9df7713fc5b988b3f5f83179","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.527,"rawValue":"0.527","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=17;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-e3819cda4be484343d54b850","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_xhigh","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.616,"rawValue":"0.616","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=10;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-54158735bed7bab4fcb35fb6","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.58,"rawValue":"0.58","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=13;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d24b4c21c8cc6d33ae4e62fb","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.531,"rawValue":"0.531","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=16;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3c4c10e0ecf7604e6d16951e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.623,"rawValue":"0.623","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=7;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-33303bc7beca25c898c9206d","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.561,"rawValue":"0.561","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=14;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-36e4ee07a90407eff2a552d4","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_xhigh","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.594,"rawValue":"0.594","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=9;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-607cb45d909e9129736624ff","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.667,"rawValue":"0.667","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=32;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-814deee0156b84edfcdcb256","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.628,"rawValue":"0.628","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=41;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-302fc9375aafe295dc104aa1","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.7000000000000001,"rawValue":"0.7000000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=30;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-e1c969cb7899497bc040dc3b","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.643,"rawValue":"0.643","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=39;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0f51c3636a0ab9d98e6db757","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_xhigh","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.693,"rawValue":"0.693","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=31;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-ac1c6b71c3f1f92362f85ed6","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-02-17","metricId":"Score","value":0.488,"rawValue":"0.488","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=22;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-330c0a7959e7b1588ab1835c","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-02-17","metricId":"Score","value":0.49,"rawValue":"0.49","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=20;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-505e0caa9554ca8a3a5df997","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-02-17","metricId":"Score","value":0.46,"rawValue":"0.46","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=25;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-fb294cf8b6af1eb111aa9757","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-30","metricId":"Score","value":0.569,"rawValue":"0.569","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=52;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-38f5af9b2af8291284278874","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-30","metricId":"Score","value":0.47700000000000004,"rawValue":"0.47700000000000004","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=65;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-db367adfb7dd736266669201","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_max","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-30","metricId":"Score","value":0.615,"rawValue":"0.615","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=27;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-bc6292596c4648cab0b8e5a8","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-30","metricId":"Score","value":0.524,"rawValue":"0.524","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=59;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b0eea32127933f627df3c7ba","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_xhigh","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-30","metricId":"Score","value":0.5870000000000001,"rawValue":"0.5870000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=28;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-ba38abe1e4ac2b29e97873c1","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_unknown","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-05-19","metricId":"Score","value":0.498,"rawValue":"0.498","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=19;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e7506baa5e04bea25f7274ff","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.535,"rawValue":"0.535","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=57;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-df19af12ebd2e398cfb78385","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.474,"rawValue":"0.474","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=66;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-0939581e4df4b286af90e7d8","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.512,"rawValue":"0.512","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=60;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-7ef937f8a04a893435235675","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-08-13","metricId":"Score","value":0.616,"rawValue":"0.616","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=43;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d19e4b9e199bac6259fc924a","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-08-13","metricId":"Score","value":0.538,"rawValue":"0.538","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=56;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4e4fc7c747a6601354f019fc","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-08-13","metricId":"Score","value":0.59,"rawValue":"0.59","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=50;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-03697f8c8056c8675ea536ed","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-16","metricId":"Score","value":0.515,"rawValue":"0.515","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=18;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-852df93f5f7ec6b3394577bf","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-16","metricId":"Score","value":0.55,"rawValue":"0.55","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=15;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-04faf7dde8d20515df63cc29","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.584,"rawValue":"0.584","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=8;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-9873a1eff70d434ee8062c83","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.466,"rawValue":"0.466","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=21;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3ccebc171b86b1813265c447","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.538,"rawValue":"0.538","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=12;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-38d00b5a1dbe3d0b38985f49","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.584,"rawValue":"0.584","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=6;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-f622390ad3c45f722282ad9c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.568,"rawValue":"0.568","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=53;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-5705feeac2ee7f037e24817b","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.376,"rawValue":"0.376","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=68;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-03cffd860ba19482f4690675","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.611,"rawValue":"0.611","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=44;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-0114d68764f1a6afcfe8cf11","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.47700000000000004,"rawValue":"0.47700000000000004","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=64;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-faec571b28a64e5dc4693775","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_xhigh","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.5770000000000001,"rawValue":"0.5770000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=51;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d0951e8a6a2dfea1bd2e4637","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.635,"rawValue":"0.635","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=40;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8be51d18d70a927bdd7187d6","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.526,"rawValue":"0.526","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=58;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-0e11c2a2fcb9462a89fac733","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.672,"rawValue":"0.672","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=29;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-c9b6965b0dbbbc34851c22c0","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.6,"rawValue":"0.6","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=47;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-6cf71149871a85664ddcb8d1","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.645,"rawValue":"0.645","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=38;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-8f02044307f619f42d98af64","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.542,"rawValue":"0.542","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=55;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-411579c278d01373f3b663e9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.469,"rawValue":"0.469","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=67;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-e3a68530577e4f285643f889","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.649,"rawValue":"0.649","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=37;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c58cd2564392b72ceb1f38aa","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.503,"rawValue":"0.503","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=62;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ba678e10845b38d3125fdb9d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.5920000000000001,"rawValue":"0.5920000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=49;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b181d2e4ad41a54e0e1a959b","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-08-12","metricId":"Score","value":0.6990000000000001,"rawValue":"0.6990000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=34;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-8ed5fe0ef2c6406d68891e10","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-08-12","metricId":"Score","value":0.61,"rawValue":"0.61","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=45;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-c168193b74508e556625aa52","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_medium","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-08-12","metricId":"Score","value":0.6709999999999999,"rawValue":"0.6709999999999999","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=35;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b6e28ea15621975307103f18","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_xhigh","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-08-12","metricId":"Score","value":0.708,"rawValue":"0.708","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=33;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-acb0cbfac40acf4e8fa5502c","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-01-27","metricId":"Score","value":0.319,"rawValue":"0.319","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=26;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e0e649c38aa698a45d60ab2a","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-04-20","metricId":"Score","value":0.47600000000000003,"rawValue":"0.47600000000000003","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=24;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-dc1ce9b1d9bd5f2275402742","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-06-12","metricId":"Score","value":0.49700000000000005,"rawValue":"0.49700000000000005","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=63;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9be92eea5cfc86f902571533","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_high","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-16","metricId":"Score","value":0.5970000000000001,"rawValue":"0.5970000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=48;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b19f9f84708e9869bdd63155","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_low","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-16","metricId":"Score","value":0.505,"rawValue":"0.505","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=61;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-44a00db99d686293ecca6a7b","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-cursorbench_external","benchmarkVersion":"2026-07-16","metricId":"Score","value":0.608,"rawValue":"0.608","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=46;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-3e71e8bcfaecd53f4e8a41d4","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-05-28","metricId":"Average score","value":35.099999999999994,"rawValue":"0.351","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=42;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://drb.futuresearch.ai"},{"id":"pub-cd763ee278375f6f95a64544","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-10-15","metricId":"Average score","value":45.5,"rawValue":"0.455","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=30;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-458a24fae7d0bb8f67049e57","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_2K","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-05-22","metricId":"Average score","value":46.8,"rawValue":"0.46799999999999997","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=27;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://drb.futuresearch.ai"},{"id":"pub-ebae4740211513c83abfc630","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_high","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-11-24","metricId":"Average score","value":54.79999999999999,"rawValue":"0.5479999999999999","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=4;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b6463caf36b5116a2d3a6c34","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-11-24","metricId":"Average score","value":53.7,"rawValue":"0.537","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=6;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e9ff875369e0f7b362ed5d4f","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_high","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-02-05","metricId":"Average score","value":55.3,"rawValue":"0.5529999999999999","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=2;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-08a3be5da7b9620ea600c0a5","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-02-05","metricId":"Average score","value":51.4,"rawValue":"0.514","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=9;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-c75950b4f5b2e13c0cf44622","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_medium","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-02-05","metricId":"Average score","value":53.2,"rawValue":"0.532","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=7;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ff674b7731f1c913f6362284","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_high","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-05-28","metricId":"Average score","value":50.2,"rawValue":"0.502","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=11;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1b5ca72526bc78196c38fa7b","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-05-28","metricId":"Average score","value":49.3,"rawValue":"0.493","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=15;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-da431f1e00acf39ff8243c7c","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_medium","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-05-28","metricId":"Average score","value":47.4,"rawValue":"0.474","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=25;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-2d0b92b94c80b63450836b1b","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_2K","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-09-29","metricId":"Average score","value":52.6,"rawValue":"0.526","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=8;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://drb.futuresearch.ai"},{"id":"pub-2ba445b1dfd4b9074bab019b","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-09-29","metricId":"Average score","value":47.5,"rawValue":"0.47500000000000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=24;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e7a22ad6cac30fc1532c2335","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_high","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-02-17","metricId":"Average score","value":54.900000000000006,"rawValue":"0.549","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=3;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b858e5bf0e2cbe3b54807b2e","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-02-17","metricId":"Average score","value":50.4,"rawValue":"0.504","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=10;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-e82138e6c38f8c9b90780f47","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-06-17","metricId":"Average score","value":41.5,"rawValue":"0.41500000000000004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=36;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c50591a9fe6ad9edad0d1243","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_high","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-02-19","metricId":"Average score","value":47.8,"rawValue":"0.478","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=23;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a417fa771310d78f1d0cfaad","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-02-19","metricId":"Average score","value":44.5,"rawValue":"0.445","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=32;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-31a24cd611aa59bfb4259b90","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-08-07","metricId":"Average score","value":48.1,"rawValue":"0.48100000000000004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=21;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4,"evidenceUrl":"https://drb.futuresearch.ai"},{"id":"pub-77e97b1827e62c680ba4403c","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-08-07","metricId":"Average score","value":49.60000000000001,"rawValue":"0.49600000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=13;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3,"evidenceUrl":"https://drb.futuresearch.ai"},{"id":"pub-80a9d616f60b5568d2115d60","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-08-07","metricId":"Average score","value":48.6,"rawValue":"0.48600000000000004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=19;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://drb.futuresearch.ai"},{"id":"pub-d529a252b7d36cd4cb231514","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_minimal","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-08-07","metricId":"Average score","value":48.9,"rawValue":"0.489","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=17;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://drb.futuresearch.ai"},{"id":"pub-10bf061e3dd1fa82492e5185","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-11-13","metricId":"Average score","value":42.8,"rawValue":"0.428","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=35;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a07a7147843d1cdd4ee26108","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2025-12-11","metricId":"Average score","value":41.1,"rawValue":"0.41100000000000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=37;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7d7548a1e648458099da87e4","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-03-05","metricId":"Average score","value":35.1,"rawValue":"0.35100000000000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=41;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b5b546d17a9000bab6bcec0f","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-03-17","metricId":"Average score","value":36.3,"rawValue":"0.363","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=40;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8db43b102d4c9e4e6eda67f6","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-04-23","metricId":"Average score","value":54.0,"rawValue":"0.54","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=5;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6ac87e66399417da4c08bf0a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_low","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-04-23","metricId":"Average score","value":48.7,"rawValue":"0.48700000000000004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=18;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-51dc2973fad304c36fb4d155","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_medium","benchmarkId":"epoch-deepresearchbench_external","benchmarkVersion":"2026-04-23","metricId":"Average score","value":49.60000000000001,"rawValue":"0.49600000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=14;column=Average score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-4006cfb8beeb423641c60d68","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-09","metricId":"Pass@1","value":68.6046511627907,"rawValue":"0.6860465116279071","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=8;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-c422ee4a9ddc3f9abb7d1048","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_low","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-09","metricId":"Pass@1","value":59.58429561200924,"rawValue":"0.5958429561200924","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=45;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ac5c103ae7a241051afb9489","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-09","metricId":"Pass@1","value":69.72477064220183,"rawValue":"0.6972477064220184","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=5;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-e1b549213bccbe346e48de55","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_medium","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-09","metricId":"Pass@1","value":65.36697247706422,"rawValue":"0.6536697247706422","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=44;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-49d3d2a30c32b4815c768c4a","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_xhigh","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-09","metricId":"Pass@1","value":69.91150442477876,"rawValue":"0.6991150442477876","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=4;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-feefd05ef1a070c7a871977e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-05-28","metricId":"Pass@1","value":51.76991150442478,"rawValue":"0.5176991150442478","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=21;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c2b3f8e300ecdfd392838c04","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_low","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-05-28","metricId":"Pass@1","value":40.79822616407982,"rawValue":"0.4079822616407982","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=29;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-1051a00af89b1d3a0bda5530","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-05-28","metricId":"Pass@1","value":58.97435897435898,"rawValue":"0.5897435897435898","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=14;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-b0f4c707b0b30a21732b03dc","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_medium","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-05-28","metricId":"Pass@1","value":48.67256637168141,"rawValue":"0.48672566371681414","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=24;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-17887d18f4d15476f75b18b4","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_xhigh","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-05-28","metricId":"Pass@1","value":54.36241610738255,"rawValue":"0.5436241610738255","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=16;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-66ff88e44a1fd79ca24c61e3","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-24","metricId":"Pass@1","value":72.82850779510022,"rawValue":"0.7282850779510023","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=41;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-fef2a4a08594905904be386d","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_low","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-24","metricId":"Pass@1","value":58.12917594654788,"rawValue":"0.5812917594654788","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=46;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fe8c81754ce28181adcd6113","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-24","metricId":"Pass@1","value":73.64864864864865,"rawValue":"0.7364864864864865","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=39;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-01a745c8ae82987f1dcf33eb","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_medium","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-24","metricId":"Pass@1","value":68.90380313199105,"rawValue":"0.6890380313199106","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=42;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-4b1f6639e2ecab355a42c83d","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_xhigh","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-24","metricId":"Pass@1","value":73.15436241610739,"rawValue":"0.7315436241610739","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=40;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-987c8c783b5684c335f69acf","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-02-17","metricId":"Pass@1","value":29.933481152993345,"rawValue":"0.29933481152993346","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=33;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-00eb0fd0b539700fb70fe3b7","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-30","metricId":"Pass@1","value":48.23008849557522,"rawValue":"0.4823008849557522","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=25;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-6bf522dcf76410c351a8199f","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_low","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-30","metricId":"Pass@1","value":30.51224944320713,"rawValue":"0.3051224944320713","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=51;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-388b084c2b544ae00cef4e66","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_max","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-30","metricId":"Pass@1","value":53.84615384615385,"rawValue":"0.5384615384615384","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=18;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-efa238ed92a5f4584024d50b","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_medium","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-30","metricId":"Pass@1","value":39.77777777777778,"rawValue":"0.3977777777777778","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=49;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a8b1d2dae4ceee9fe4aedfc6","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_xhigh","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-30","metricId":"Pass@1","value":49.667405764966745,"rawValue":"0.49667405764966743","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=23;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-c089a9a3df1062b31994a310","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-02-19","metricId":"Pass@1","value":11.725663716814159,"rawValue":"0.1172566371681416","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=36;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-846a602a2c2f1267f34630a7","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-05-19","metricId":"Pass@1","value":36.06194690265487,"rawValue":"0.3606194690265487","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=57;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c0850a5da54b97511a19f7f3","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_medium","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-05-19","metricId":"Pass@1","value":37.389380530973455,"rawValue":"0.37389380530973454","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=50;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9268916a0092830c60e48793","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-21","metricId":"Pass@1","value":46.68141592920354,"rawValue":"0.4668141592920354","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=48;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c54120a92c4d1ff0beb83f8c","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-08-13","metricId":"Pass@1","value":65.2654867256637,"rawValue":"0.6526548672566371","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=59;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8be0a9d2c82a4cecbe8e9b57","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_low","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-08-13","metricId":"Pass@1","value":53.76106194690266,"rawValue":"0.5376106194690266","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=60;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-984e672ad8c3d0bdac675773","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_medium","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-08-13","metricId":"Pass@1","value":65.48672566371681,"rawValue":"0.6548672566371682","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=58;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-34768d821cafec20ffed8ce7","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-16","metricId":"Pass@1","value":36.283185840707965,"rawValue":"0.36283185840707965","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=30;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ef20c9f778d375b7e9d27640","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-16","metricId":"Pass@1","value":43.777777777777786,"rawValue":"0.4377777777777778","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=28;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b13ff96f4fd4df8970358612","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3_max","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-08-14","metricId":"Pass@1","value":68.95787139689578,"rawValue":"0.6895787139689579","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=61;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1ff031f17e0ca00109b2b59b","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-03-05","metricId":"Pass@1","value":51.76991150442478,"rawValue":"0.5176991150442478","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=22;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3e26486b908330a7b3e04510","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-04-23","metricId":"Pass@1","value":64.38053097345133,"rawValue":"0.6438053097345133","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=11;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-dac87db8c75127ce84e6a384","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_low","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-04-23","metricId":"Pass@1","value":26.991150442477874,"rawValue":"0.26991150442477874","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=34;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-e522eb3ea96be2633f714fbd","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_medium","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-04-23","metricId":"Pass@1","value":53.98230088495575,"rawValue":"0.5398230088495575","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=17;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-57ecee3160942a344e8fca07","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-04-23","metricId":"Pass@1","value":67.03539823008849,"rawValue":"0.6703539823008849","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=10;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-28a2ae7bb1e7d4c0d767103c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":44.24778761061947,"rawValue":"0.4424778761061947","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=27;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-26470f3000a2bdf6bdcb248e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_low","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":1.5486725663716814,"rawValue":"0.015486725663716814","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=38;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-bb2638a6d5816563ca92d7eb","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":67.1875,"rawValue":"0.671875","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=9;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7d71cfd019ca25326cd215aa","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_medium","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":11.283185840707963,"rawValue":"0.11283185840707964","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=37;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-1e9e65475e542a06dfa1dd3e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_xhigh","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":56.85840707964602,"rawValue":"0.5685840707964602","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=15;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-5826ef81cdd691d50f75543a","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":69.40133037694012,"rawValue":"0.6940133037694013","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=7;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-3ba37e445b3f805b0e298b7b","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_low","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":45.35398230088495,"rawValue":"0.45353982300884954","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=26;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-241cf6362fb6190218f5a9ea","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":72.66666666666667,"rawValue":"0.7266666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=2;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-270b3dc7d6300aba2c04117e","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_medium","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":61.06194690265486,"rawValue":"0.6106194690265486","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=12;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-a69d821c31df6e781fad66cd","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":70.73170731707317,"rawValue":"0.7073170731707318","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=3;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3301d03829d4efde8d520b37","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":53.76106194690266,"rawValue":"0.5376106194690266","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=19;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-eac37568f1c8a9bbf330c79f","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_low","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":24.053452115812917,"rawValue":"0.24053452115812918","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=35;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e32ee58097f7c552f2ed68d2","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":69.62305986696231,"rawValue":"0.6962305986696231","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=6;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b2cc7adea37d644577ccf7c0","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_medium","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":35.11111111111111,"rawValue":"0.3511111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=31;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-4a2880700ff84412f5ced15d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-09","metricId":"Pass@1","value":60.17699115044248,"rawValue":"0.6017699115044248","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=13;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-fa49913d440bf7532cdbc150","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-08","metricId":"Pass@1","value":53.76106194690266,"rawValue":"0.5376106194690266","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=47;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-673b43b063f3a2e7084d6185","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_high","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-08-12","metricId":"Pass@1","value":65.18847006651885,"rawValue":"0.6518847006651884","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=54;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-f1763c8bb34340c1e1d00532","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_low","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-08-12","metricId":"Pass@1","value":41.648106904231625,"rawValue":"0.41648106904231624","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=56;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bfe347ccce01c5792c70493f","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_medium","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-08-12","metricId":"Pass@1","value":67.47787610619469,"rawValue":"0.6747787610619469","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=52;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-bdf3565f4da9e7193935c671","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_xhigh","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-08-12","metricId":"Pass@1","value":66.74057649667405,"rawValue":"0.6674057649667405","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=53;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-956ce259f15ae719e29cb55d","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-06-12","metricId":"Pass@1","value":30.53097345132743,"rawValue":"0.3053097345132743","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=32;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-463e5f1dd24cdfd06d9f2250","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-07-16","metricId":"Pass@1","value":68.51441241685144,"rawValue":"0.6851441241685144","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=43;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4c0465286098c6cdd74fc847","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max_xhigh","benchmarkId":"epoch-deepswe_external","benchmarkVersion":"2026-08-02","metricId":"Pass@1","value":57.46102449888642,"rawValue":"0.5746102449888641","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=55;column=Pass@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-754e717dd113a0303340ee62","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2026-06-09","metricId":"Accuracy","value":39.28,"rawValue":"0.39280000000000004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=44;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a241445c4a9f02a94783c3e9","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_unknown","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2025-05-22","metricId":"Accuracy","value":5.57,"rawValue":"0.055700000000000006","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=22;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a39bba733ba5df1eaddbaf74","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy","value":11.91,"rawValue":"0.1191","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=8;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4cc5fe33efa8f04d576e49a0","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy","value":6.84,"rawValue":"0.0684","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=17;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-7fa89aca60eb77dbfecf21da","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy","value":7.6,"rawValue":"0.076","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=15;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6355637b9bceec706cf9ca45","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_xhigh","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2026-05-28","metricId":"Accuracy","value":23.51,"rawValue":"0.23510000000000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=47;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-34aff24e9a6786f13dc30e46","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy","value":5.9966,"rawValue":"0.059966","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=20;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6d71036cd95d4c36e3dfd7e1","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2026-02-19","metricId":"Accuracy","value":36.78,"rawValue":"0.3678","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=3;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9c622144d4ce69a7fd9ccf2b","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2026-05-19","metricId":"Accuracy","value":25.41,"rawValue":"0.2541","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=46;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1f757a275822dad0ffe30496","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_unknown","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":10.470000000000002,"rawValue":"0.10470000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=11;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c289d5dba3c4838437e74881","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":8.1926,"rawValue":"0.08192600000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=14;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d7ab36b3539a82712c5bb4fc","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_none","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2025-11-13","metricId":"Accuracy","value":1.94,"rawValue":"0.0194","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=34;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-fe6fd95a0dba3304cbd9ff21","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2025-11-13","metricId":"Accuracy","value":11.23,"rawValue":"0.11230000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=10;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1b4b1852fba9bf5958d69022","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy","value":10.39,"rawValue":"0.1039","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=12;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5c9240521729fd44def97079","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2026-03-05","metricId":"Accuracy","value":15.960000000000003,"rawValue":"0.15960000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=6;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6a4c58205965c624f35b12b0","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_unknown","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2026-03-05","metricId":"Accuracy","value":23.82,"rawValue":"0.2382","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=2;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ccc2daba93c2478d2b11b2c4","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_high","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":37.12,"rawValue":"0.3712","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=45;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-26fa9c6fc4d43e94aae55752","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-enigma_eval_external","benchmarkVersion":"2026-01-27","metricId":"Accuracy","value":3.38,"rawValue":"0.0338","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=26;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6572ddc1e5c20a476e2822b7","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-05-28","metricId":"ECI Score","value":141.97,"rawValue":"141.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=223;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-8b2ec6dec50bda65f974b705","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":144.99,"rawValue":"144.99","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=489;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ceb6250906df6b68c6d42947","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":144.99,"rawValue":"144.99","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=843;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-534f3d3a3230d86bbe35dba9","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp_thinking","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":144.99,"rawValue":"144.99","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=490;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-d695183ed177e92bfc15c03e","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":144.99,"rawValue":"144.99","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=800;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e782cc15081f151db780967a","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-12","metricId":"ECI Score","value":146.78,"rawValue":"146.78","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=542;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0beb50bd9675fad6f8b3d343","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMax-M2.7","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-18","metricId":"ECI Score","value":146.0,"rawValue":"146.0","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=393;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-88930990717599be62b68967","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-01","metricId":"ECI Score","value":147.76,"rawValue":"147.76","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=26;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b7bee9d92a02e9ac2eaf8d8b","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"chutes/DeepSeek-R1-0528","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-05-28","metricId":"ECI Score","value":141.97,"rawValue":"141.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=833;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-53df77c27e301362df156951","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-09","metricId":"ECI Score","value":162.48,"rawValue":"162.48","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=513;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-1f261f3995f874979fdf57dd","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-09","metricId":"ECI Score","value":162.48,"rawValue":"162.48","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=13;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-7b31ab4da11ebf6e9a0a65f4","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-09","metricId":"ECI Score","value":162.48,"rawValue":"162.48","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=101;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-e0a2829f9464845edc6a0555","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-09","metricId":"ECI Score","value":162.48,"rawValue":"162.48","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=102;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3b9447d2da8ab000c16eebd3","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-09","metricId":"ECI Score","value":162.48,"rawValue":"162.48","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=351;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-949990cf45894d2c923c7899","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-09","metricId":"ECI Score","value":162.48,"rawValue":"162.48","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=471;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-d4a6d1485050f8599e821238","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-09","metricId":"ECI Score","value":162.48,"rawValue":"162.48","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=18;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-cf4f8b797a03d6b93cb81fb0","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-10-15","metricId":"ECI Score","value":142.99,"rawValue":"142.99","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=25;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d13120a9d3d4b1a534a0dbb9","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_16K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-10-15","metricId":"ECI Score","value":142.99,"rawValue":"142.99","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=637;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-57b0ce26b1c24ae5dc4a6e03","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_1K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-10-15","metricId":"ECI Score","value":142.99,"rawValue":"142.99","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=654;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-02293efda7b024d61fd47b50","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_32K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-10-15","metricId":"ECI Score","value":142.99,"rawValue":"142.99","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=137;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-61737660e0801984be6c67a7","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_8K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-10-15","metricId":"ECI Score","value":142.99,"rawValue":"142.99","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=651;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-3e3ec8f8d75e420e4e336280","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-10-15","metricId":"ECI Score","value":142.99,"rawValue":"142.99","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=811;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-b896cea46e31db88aa08a8cb","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-10-15","metricId":"ECI Score","value":142.99,"rawValue":"142.99","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=392;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-6938b7eda5b2d8ff118ee30a","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-05-22","metricId":"ECI Score","value":143.02,"rawValue":"143.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=200;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-2e9decf668f771da5f8a8931","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_12K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-05-22","metricId":"ECI Score","value":143.02,"rawValue":"143.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=789;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-46a3d8486175ed24c6eccc25","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_16K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-05-22","metricId":"ECI Score","value":143.02,"rawValue":"143.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=245;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-18b584a306ec03e013ee62bd","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_1K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-05-22","metricId":"ECI Score","value":143.02,"rawValue":"143.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=649;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-65b5dcae4d6d41cd66faee60","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_27K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-05-22","metricId":"ECI Score","value":143.02,"rawValue":"143.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=232;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-f30ba4b63a46b46a073d6dd7","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_2K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-05-22","metricId":"ECI Score","value":143.02,"rawValue":"143.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=808;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-315a46e328b49855c16da22d","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_32K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-05-22","metricId":"ECI Score","value":143.02,"rawValue":"143.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=495;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-c885469151dd11b185f75335","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_8K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-05-22","metricId":"ECI Score","value":143.02,"rawValue":"143.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=644;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-617621f23f4ff1a4d9549d5f","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-05-22","metricId":"ECI Score","value":143.02,"rawValue":"143.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=335;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-bdd1fe10a480d906b2db8ae6","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-24","metricId":"ECI Score","value":149.97,"rawValue":"149.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=100;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-d8140465a6b5a4878f531ea0","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_128K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-24","metricId":"ECI Score","value":149.97,"rawValue":"149.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=163;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-c3b30834ffe56d7172f22c22","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_16K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-24","metricId":"ECI Score","value":149.97,"rawValue":"149.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=225;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-9d89c5b541388cd39aab91d6","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_32K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-24","metricId":"ECI Score","value":149.97,"rawValue":"149.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=179;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-da22570db25682d30e7d8fe9","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_48K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-24","metricId":"ECI Score","value":149.97,"rawValue":"149.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=131;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-afd17daecc732362a69e98de","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_64K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-24","metricId":"ECI Score","value":149.97,"rawValue":"149.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=623;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-f26b5885f1c577deaaf355d8","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_8K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-24","metricId":"ECI Score","value":149.97,"rawValue":"149.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=628;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-550ead7548405150aa6311ea","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-24","metricId":"ECI Score","value":149.97,"rawValue":"149.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=803;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-bcc6ca1eb8e8d53253515a8d","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-24","metricId":"ECI Score","value":149.97,"rawValue":"149.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=804;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-5446b3f0eaff0ed6ac2cc315","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-24","metricId":"ECI Score","value":149.97,"rawValue":"149.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=332;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":10},{"id":"pub-fcd7854f7e460065f9707a15","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":154.98,"rawValue":"154.98","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=109;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-db27f9b616a690c75c80d505","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_120K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":154.98,"rawValue":"154.98","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=192;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c2bef93bac0b283389614d19","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_32K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":154.98,"rawValue":"154.98","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=199;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-8f2a0b8aade3cea69c49d277","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_64K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":154.98,"rawValue":"154.98","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=198;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-8beae8bc03560ba16581cf10","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":154.98,"rawValue":"154.98","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=329;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-6b154d392ffeb2b5b47bc0c5","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":154.98,"rawValue":"154.98","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=110;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-08eb78129c8efc565ba68d48","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":154.98,"rawValue":"154.98","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=83;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-34f6eaf0477fe6cfafe583a3","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":154.98,"rawValue":"154.98","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=805;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-40f3b31b29569dbb826ba8ae","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":154.98,"rawValue":"154.98","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=331;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-9790b2fbe2d9e8fc1676ba6a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-16","metricId":"ECI Score","value":155.93,"rawValue":"155.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=108;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-662fcb595906225774c4d132","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-16","metricId":"ECI Score","value":155.93,"rawValue":"155.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=8;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-68cde21ff1525257f62b56e5","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-16","metricId":"ECI Score","value":155.93,"rawValue":"155.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=151;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-cf3b29521a9cc3c789dd2fe6","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-16","metricId":"ECI Score","value":155.93,"rawValue":"155.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=99;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-bdedd7a0836bccf3f011a73d","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-16","metricId":"ECI Score","value":155.93,"rawValue":"155.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=345;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-ce7d795a3cdfe58d13f98867","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-16","metricId":"ECI Score","value":155.93,"rawValue":"155.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=476;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ea87d3749c793bc1ec49b209","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-16","metricId":"ECI Score","value":155.93,"rawValue":"155.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=183;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f87b0509e00525462865096d","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_24K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-28","metricId":"ECI Score","value":157.73,"rawValue":"157.73","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=601;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-dda95f98f44d4349f98dcd3b","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-28","metricId":"ECI Score","value":157.73,"rawValue":"157.73","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=11;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-fd27227ec5eb96acd128a9bd","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-28","metricId":"ECI Score","value":157.73,"rawValue":"157.73","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=98;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-14ba5ee59b66f51c36c040f9","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-28","metricId":"ECI Score","value":157.73,"rawValue":"157.73","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=32;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-33b17f6df88ee059df117c0e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-28","metricId":"ECI Score","value":157.73,"rawValue":"157.73","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=344;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-1cbae1b3b56a9e70444a976e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-28","metricId":"ECI Score","value":157.73,"rawValue":"157.73","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=97;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-7b90eaf89582840a9e9b82f9","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-28","metricId":"ECI Score","value":157.73,"rawValue":"157.73","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=328;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-47aa7c46d3bcc7846b21219b","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-28","metricId":"ECI Score","value":157.73,"rawValue":"157.73","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=128;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-c2e3f9878b268bab9e975502","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-24","metricId":"ECI Score","value":161.59,"rawValue":"161.59","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=96;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-e3c785c3b2b3734ecee711a7","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-24","metricId":"ECI Score","value":161.59,"rawValue":"161.59","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=349;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-766ce1d608af606c95c7028c","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-24","metricId":"ECI Score","value":161.59,"rawValue":"161.59","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=94;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-41b5d17bff39496f5669f8cf","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-24","metricId":"ECI Score","value":161.59,"rawValue":"161.59","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=17;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-5be46d4e95ccf47383aeb7a3","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-24","metricId":"ECI Score","value":161.59,"rawValue":"161.59","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=353;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-d6d505d1c00a1424df423759","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-24","metricId":"ECI Score","value":161.59,"rawValue":"161.59","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=470;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-a5a39ba6c195aa2fbe58b0b9","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-24","metricId":"ECI Score","value":161.59,"rawValue":"161.59","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=348;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-28df86e0cb659a1702211d8c","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":146.93,"rawValue":"146.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=20;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":10},{"id":"pub-a23eb17ce804037f984adcf8","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_12K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":146.93,"rawValue":"146.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=792;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-db1f7b183c6f4dbefd5807a6","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_16K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":146.93,"rawValue":"146.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=227;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-dd4ec918ca5c41fcba58c2ab","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_1K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":146.93,"rawValue":"146.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=643;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-6eb7a1bbcfc4896095935a8c","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_2K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":146.93,"rawValue":"146.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=806;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-8c7c16e3b5da148501a8a174","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":146.93,"rawValue":"146.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=176;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-fdc92e6db7790c5c3a9cec94","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_48K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":146.93,"rawValue":"146.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=129;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-787ff5af1c429a6081db60c6","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_59K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":146.93,"rawValue":"146.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=218;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-92f26a9556a0447df3ca265c","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_8K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":146.93,"rawValue":"146.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=632;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-6af39b1515ac3bc29f7fcb94","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":146.93,"rawValue":"146.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=807;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-12d1460273999d392df4dd97","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-29","metricId":"ECI Score","value":146.93,"rawValue":"146.93","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=333;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":11},{"id":"pub-0f10459a5bad426ec0679839","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-17","metricId":"ECI Score","value":152.16,"rawValue":"152.16","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=106;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":7},{"id":"pub-94f3de2856145d5c9d8c59a6","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_16K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-17","metricId":"ECI Score","value":152.16,"rawValue":"152.16","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=190;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-31fc3409ea0a6b1fd19cf59e","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_32K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-17","metricId":"ECI Score","value":152.16,"rawValue":"152.16","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=191;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-016556b11a3fe7ba8c2a4237","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-17","metricId":"ECI Score","value":152.16,"rawValue":"152.16","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=19;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-e48cf957e16ca268ca5be186","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-17","metricId":"ECI Score","value":152.16,"rawValue":"152.16","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=107;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-df2e40898442e44b608e4362","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-17","metricId":"ECI Score","value":152.16,"rawValue":"152.16","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=95;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-31281a414efbc10f0a4e0a56","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-17","metricId":"ECI Score","value":152.16,"rawValue":"152.16","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=155;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-5acff9424685b7597b6f40a5","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-17","metricId":"ECI Score","value":152.16,"rawValue":"152.16","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=474;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-86b705da117ee468908eba99","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-30","metricId":"ECI Score","value":155.87,"rawValue":"155.87","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=113;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-835a3976d85bfd3cbc73e524","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_16K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-30","metricId":"ECI Score","value":155.87,"rawValue":"155.87","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=598;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-c3fad886c767c1b8cba9da2c","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-30","metricId":"ECI Score","value":155.87,"rawValue":"155.87","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=358;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ddb562303e0f80bb3c070f01","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-30","metricId":"ECI Score","value":155.87,"rawValue":"155.87","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=367;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1104f0ccb849742b4e901261","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-30","metricId":"ECI Score","value":155.87,"rawValue":"155.87","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=93;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-161a7106d1e9f9f979881f20","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-30","metricId":"ECI Score","value":155.87,"rawValue":"155.87","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=363;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-2af9313dac1869bde8f71ea7","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-30","metricId":"ECI Score","value":155.87,"rawValue":"155.87","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=330;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-87ea3047a88fea11a4a5559b","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-30","metricId":"ECI Score","value":155.87,"rawValue":"155.87","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=160;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-8e178af3f6b28f21d9f54b44","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-chat","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-01","metricId":"ECI Score","value":146.47,"rawValue":"146.47","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=138;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-5a162232e5555b65b0feb9b9","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-31","metricId":"ECI Score","value":154.35,"rawValue":"154.35","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=619;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-7c4a0daa9e5cc4598e651681","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-31","metricId":"ECI Score","value":154.35,"rawValue":"154.35","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=621;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-cd9bfef788b3b78fab4e79e4","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-31","metricId":"ECI Score","value":154.35,"rawValue":"154.35","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=117;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-f32d342d646ba39b57e56cb4","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-31","metricId":"ECI Score","value":154.35,"rawValue":"154.35","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=659;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-00e17d90225bee0d47798af9","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-31","metricId":"ECI Score","value":154.35,"rawValue":"154.35","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=529;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-00c5f400056dabaa1f41d8fb","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-13","metricId":"ECI Score","value":155.18,"rawValue":"155.18","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=618;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-64f5a921826103511135936c","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-13","metricId":"ECI Score","value":155.18,"rawValue":"155.18","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=616;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-7224d21926a9f1e331192b9b","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-13","metricId":"ECI Score","value":155.18,"rawValue":"155.18","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=3;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-691b4c67ff0019f0f86dc9e9","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-13","metricId":"ECI Score","value":155.18,"rawValue":"155.18","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=658;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-d7cdb5411ecb89910647cb28","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-13","metricId":"ECI Score","value":155.18,"rawValue":"155.18","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=533;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8bc0e0b4574a3b29cf7d87cc","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-24","metricId":"ECI Score","value":149.24,"rawValue":"149.24","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=82;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-94883d50bfcfaec0321b146f","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-24","metricId":"ECI Score","value":149.24,"rawValue":"149.24","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=165;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-33f7be78c2d0b6b15f9132fe","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-24","metricId":"ECI Score","value":149.24,"rawValue":"149.24","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=81;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-1077f56f2c256f053cca051c","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-24","metricId":"ECI Score","value":149.24,"rawValue":"149.24","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=610;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-c3ad8abcb8ee824c22fca831","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek/deepseek-v3.2","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-01","metricId":"ECI Score","value":146.47,"rawValue":"146.47","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=629;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-194362729390da3e6185b307","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"fireworks/deepseek-v3p2","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-01","metricId":"ECI Score","value":146.47,"rawValue":"146.47","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=208;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-9922e0b7614c5d571822bd8b","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"fireworks/kimi-k2p5","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-01-27","metricId":"ECI Score","value":148.22,"rawValue":"148.22","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=202;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e78e9cb695af6309c8538d13","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-06-17","metricId":"ECI Score","value":140.52,"rawValue":"140.52","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=209;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-229795892958f3668b3bc927","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-06-17","metricId":"ECI Score","value":145.84,"rawValue":"145.84","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=173;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-5a24767422d5a336b80cf1cc","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro_32K","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-06-17","metricId":"ECI Score","value":145.84,"rawValue":"145.84","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=638;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b779a4768e08d327aa1d29aa","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-19","metricId":"ECI Score","value":154.61,"rawValue":"154.61","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=164;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-385bdc125efa0ff52c73b7df","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-19","metricId":"ECI Score","value":154.61,"rawValue":"154.61","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=21;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-9537be9618b23e64f313b9e9","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-19","metricId":"ECI Score","value":154.61,"rawValue":"154.61","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=112;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-4831e82a39bc69c3c8d56bb8","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-19","metricId":"ECI Score","value":154.61,"rawValue":"154.61","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=111;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-ac3e79ad233658dd64123529","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-21","metricId":"ECI Score","value":145.19,"rawValue":"145.19","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=538;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8ac9a97a63366310689dba6a","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-21","metricId":"ECI Score","value":145.19,"rawValue":"145.19","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=87;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-82f071026d0e8ac65ea466a4","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-21","metricId":"ECI Score","value":145.19,"rawValue":"145.19","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=85;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-8c8ad7860b56084556597ef1","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-21","metricId":"ECI Score","value":145.19,"rawValue":"145.19","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=641;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-649459b2463d26556847beec","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_minimal","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-21","metricId":"ECI Score","value":145.19,"rawValue":"145.19","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=86;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-9957be5b750136cb9f11af8d","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-19","metricId":"ECI Score","value":154.43,"rawValue":"154.43","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=9;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-b518868ea59ef9c85f6e701c","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-19","metricId":"ECI Score","value":154.43,"rawValue":"154.43","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=88;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-3198d67a82b82f0321c8214b","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-19","metricId":"ECI Score","value":154.43,"rawValue":"154.43","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=396;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-44db85edb15428ed2c6c4def","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_minimal","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-19","metricId":"ECI Score","value":154.43,"rawValue":"154.43","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=149;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-fd14c590eb5c1254284843f3","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-19","metricId":"ECI Score","value":154.43,"rawValue":"154.43","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=347;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2f722c53fbabd6eab7b16b79","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-21","metricId":"ECI Score","value":154.14,"rawValue":"154.14","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=123;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-a495361730c4694d63b381fe","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-21","metricId":"ECI Score","value":154.14,"rawValue":"154.14","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=65;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-7dced62d9bca1b825c052316","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-21","metricId":"ECI Score","value":154.14,"rawValue":"154.14","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=364;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ed96f02f2060784f7b0a04e9","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_minimal","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-21","metricId":"ECI Score","value":154.14,"rawValue":"154.14","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=64;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-53b064837a716d72cb14dc40","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-21","metricId":"ECI Score","value":154.14,"rawValue":"154.14","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=527;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-40f8cfde24e0548a180be160","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-13","metricId":"ECI Score","value":157.12,"rawValue":"157.12","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=6;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-48575bf16bf6ed9fcdd30f95","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-13","metricId":"ECI Score","value":157.12,"rawValue":"157.12","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=362;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-be7fbf450949c7b9f99d269f","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-13","metricId":"ECI Score","value":157.12,"rawValue":"157.12","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=356;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-91aec8f7c77d5947a96fa5ae","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-13","metricId":"ECI Score","value":157.12,"rawValue":"157.12","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=514;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-951b67e96a153f26954287a0","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-22","metricId":"ECI Score","value":143.94,"rawValue":"143.94","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=206;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-764e64f0b1f5b994414a9f70","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-11","metricId":"ECI Score","value":146.2,"rawValue":"146.2","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=193;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f43bb6c136e4b720bf64bd35","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-07","metricId":"ECI Score","value":150.85,"rawValue":"150.85","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=28;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-03cc1a2ffa8124cce1114191","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-16","metricId":"ECI Score","value":152.02,"rawValue":"152.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=346;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-cf0d42e14db162ed76f894f9","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-16","metricId":"ECI Score","value":152.02,"rawValue":"152.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=161;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-976c348dadd55e2048f5134f","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-16","metricId":"ECI Score","value":152.02,"rawValue":"152.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=27;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-b4799dec78d2645764995694","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-16","metricId":"ECI Score","value":152.02,"rawValue":"152.02","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=479;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-2754bd83799b5a882b95577a","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":150.0,"rawValue":"150.0","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=122;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-312824162ec934565564054c","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":150.0,"rawValue":"150.0","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=147;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-77e46ff6e980897b68a1ed07","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":150.0,"rawValue":"150.0","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=146;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-e92731b85c05b7eba1ca88ab","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_minimal","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":150.0,"rawValue":"150.0","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=135;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f2696d8c34e968940353b2ec","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":150.0,"rawValue":"150.0","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=516;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1308e4b30101725121e5f503","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":145.91,"rawValue":"145.91","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=23;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-033b33f65fba401d4cacf6c4","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":145.91,"rawValue":"145.91","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=73;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-8f05d0fd96348e56859110b8","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":145.91,"rawValue":"145.91","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=203;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-feb3e6707a16e142cd4055c7","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_minimal","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":145.91,"rawValue":"145.91","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=72;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-03634c95a94e48fb927ac9e3","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":145.91,"rawValue":"145.91","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=567;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-25c6759ab2c3b95105e3597a","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":141.11,"rawValue":"141.11","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=24;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-1a795cc297867929acb494ff","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":141.11,"rawValue":"141.11","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=152;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-5afd4b8d6c7deba7cb8740b5","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":141.11,"rawValue":"141.11","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=231;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f1b3161373d22a1ef9b26547","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_minimal","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":141.11,"rawValue":"141.11","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=153;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-074f9ab6e2bd7dc26ce06be9","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-08-07","metricId":"ECI Score","value":141.11,"rawValue":"141.11","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=602;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-c18b90098aa7dcc7835b278d","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-13","metricId":"ECI Score","value":149.63,"rawValue":"149.63","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=194;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9a013bc9b49ed5cf2763f745","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-13","metricId":"ECI Score","value":149.63,"rawValue":"149.63","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=61;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-109d7559b0105544fd1bf706","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-13","metricId":"ECI Score","value":149.63,"rawValue":"149.63","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=226;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-6814b710503208684f25f99d","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-13","metricId":"ECI Score","value":149.63,"rawValue":"149.63","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=49;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-988b75b125cc57b8fef23253","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-11-13","metricId":"ECI Score","value":149.63,"rawValue":"149.63","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=334;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-a15d00f61f5b0e9ea39a4d89","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-11","metricId":"ECI Score","value":153.31,"rawValue":"153.31","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=105;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-98f5e4bdeea4ca183cbfcf2a","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-11","metricId":"ECI Score","value":153.31,"rawValue":"153.31","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=214;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-164d59f31a145371ba468e96","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-11","metricId":"ECI Score","value":153.31,"rawValue":"153.31","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=213;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-b4816dce4b2d8ba6d3556f46","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-11","metricId":"ECI Score","value":153.31,"rawValue":"153.31","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=154;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-73c67f487d1f7377526e3a97","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-11","metricId":"ECI Score","value":153.31,"rawValue":"153.31","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=515;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-5bfbff532544d85d4d158ac0","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-11","metricId":"ECI Score","value":153.31,"rawValue":"153.31","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=162;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-dac8c955062dd791005cb705","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-11","metricId":"ECI Score","value":155.07,"rawValue":"155.07","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=790;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c1a37b89dc3b6761cfb2c08f","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11-webapp","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-11","metricId":"ECI Score","value":155.07,"rawValue":"155.07","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=207;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-bfcd92a8d62b9c05ab754790","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-11","metricId":"ECI Score","value":155.07,"rawValue":"155.07","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=620;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-1e4ddadb27432833c8941a07","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-11","metricId":"ECI Score","value":155.07,"rawValue":"155.07","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=622;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-bf5ed940690ea0cd18e97c7b","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-11","metricId":"ECI Score","value":155.07,"rawValue":"155.07","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=168;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-5a28b74335fc6ab298896377","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":156.22,"rawValue":"156.22","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=737;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-f0add221a2983dcffc39e017","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":156.22,"rawValue":"156.22","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=188;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-848e16c0c1f26f05afdfb061","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-05","metricId":"ECI Score","value":156.22,"rawValue":"156.22","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=679;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-6031377d83302d3ae9be9613","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-05","metricId":"ECI Score","value":156.49,"rawValue":"156.49","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=30;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-39723867d286024924e22425","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-05","metricId":"ECI Score","value":156.49,"rawValue":"156.49","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=144;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-91c9bb1a7c0e1ed0522ccc67","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-05","metricId":"ECI Score","value":156.49,"rawValue":"156.49","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=142;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-65ee9a0dd096a130e5165370","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-05","metricId":"ECI Score","value":156.49,"rawValue":"156.49","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=143;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-9c058ef704c4efe12f50726a","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-05","metricId":"ECI Score","value":156.49,"rawValue":"156.49","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=475;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-21007b249146a135eec63dc0","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-05","metricId":"ECI Score","value":156.49,"rawValue":"156.49","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=132;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-58cef27096d529ef6b2f87f3","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":149.33,"rawValue":"149.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=184;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-cb37a59807028f2997fe8a49","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":149.33,"rawValue":"149.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=657;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-bd670ef8f608990ef1ef9db1","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":149.33,"rawValue":"149.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=635;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-99ad9b1b757c50c3daf6005e","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":149.33,"rawValue":"149.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=43;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-a4ff48c6193bf8f4920079c7","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":149.33,"rawValue":"149.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=528;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-fe9b863980d2e09fe6a2ad8a","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":149.33,"rawValue":"149.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=36;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e3c7735e78a677beed5f3186","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":146.29,"rawValue":"146.29","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=171;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-842289ccd6034388d46d2839","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":146.29,"rawValue":"146.29","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=79;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-eb011ba8c488ce9aacb414b4","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":146.29,"rawValue":"146.29","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=640;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-70caf568f2fe5deb988ef922","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":146.29,"rawValue":"146.29","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=77;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-e737e30d04231b5e34bd90f5","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":146.29,"rawValue":"146.29","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=606;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-000a8422ebe30d5bc4655c36","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-17","metricId":"ECI Score","value":146.29,"rawValue":"146.29","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=561;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-e245797371c34b85eb70a9c3","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05-web-app","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-05","metricId":"ECI Score","value":158.55,"rawValue":"158.55","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=187;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-ffa4993dba3f40bca2dc46c3","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-05","metricId":"ECI Score","value":158.55,"rawValue":"158.55","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=813;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f63c4638a09a94d0fe725a18","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-05","metricId":"ECI Score","value":158.55,"rawValue":"158.55","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=775;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-f8bd1f30c8a7f43a94104b97","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-05","metricId":"ECI Score","value":158.55,"rawValue":"158.55","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=167;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-fbb55a6982c563d64fb63497","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-05","metricId":"ECI Score","value":143.56,"rawValue":"143.56","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=125;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b1d365c59d4e4bcec2b58837","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-23","metricId":"ECI Score","value":161.71,"rawValue":"161.71","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=614;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8c13d86a89aec33a747b1d8e","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-23","metricId":"ECI Score","value":161.71,"rawValue":"161.71","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=786;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-a4be7fc098cac0729493a327","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-23","metricId":"ECI Score","value":161.71,"rawValue":"161.71","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=169;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-6213b0cbc0675b4579558b99","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-23","metricId":"ECI Score","value":158.67,"rawValue":"158.67","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=29;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-7b4cea1fd68f2dfefdeae332","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-23","metricId":"ECI Score","value":158.67,"rawValue":"158.67","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=46;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-8afa9da2a03845ada32475d0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-23","metricId":"ECI Score","value":158.67,"rawValue":"158.67","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=10;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-14086d9f1ca828e03a294647","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-23","metricId":"ECI Score","value":158.67,"rawValue":"158.67","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=38;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-70ec95aa871356c603806df9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-23","metricId":"ECI Score","value":158.67,"rawValue":"158.67","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=391;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-1e19f6b8bc34538b7775a8e6","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-23","metricId":"ECI Score","value":158.67,"rawValue":"158.67","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=127;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-bea3075b63f5adbe458d1d88","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":156.21,"rawValue":"156.21","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=359;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-f9fe9b58bc1fa40a94959965","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":156.21,"rawValue":"156.21","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=69;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3ddcfaaa0b1fc4e80b3c41a1","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":156.21,"rawValue":"156.21","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=16;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-083c6f97dca93ce11a038271","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":156.21,"rawValue":"156.21","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=366;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-6e47728c198608579be37f1a","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":156.21,"rawValue":"156.21","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=53;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-e839ff2cde6969170fc198c8","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":156.21,"rawValue":"156.21","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=526;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-42c958f115641d26898944ec","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":156.21,"rawValue":"156.21","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=357;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-e5660ebfd3257c5ed7578300","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":161.06,"rawValue":"161.06","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=31;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-2748cb09278d4a18436a1757","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":161.06,"rawValue":"161.06","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=44;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":8},{"id":"pub-0f84522321833abf54a65a09","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":161.06,"rawValue":"161.06","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=14;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":10},{"id":"pub-fd06215abed6837f80a684eb","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":161.06,"rawValue":"161.06","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=12;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2302bdf1cb1cd49afde092fb","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":161.06,"rawValue":"161.06","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=37;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-abee95fb8ce4467a9c473e54","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_promax","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":161.06,"rawValue":"161.06","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=158;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-b97daff572237ae4496dd324","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_prounknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":161.06,"rawValue":"161.06","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=788;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-a594f7d435a583c76cd0a899","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_proxhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":161.06,"rawValue":"161.06","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=787;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-b1fd36844ad75264a7011b61","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":161.06,"rawValue":"161.06","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=473;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-3ec1d6fb517cc971b3131363","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":161.06,"rawValue":"161.06","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=352;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":7},{"id":"pub-ee35461cf56e754f64b51724","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":158.86,"rawValue":"158.86","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=361;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-64a91034f90aa2bcbc6af4a6","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":158.86,"rawValue":"158.86","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=51;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-cae40985cb38b7ba01bbb03b","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":158.86,"rawValue":"158.86","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=15;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-d94b270e5aeeeffd3742c7f7","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":158.86,"rawValue":"158.86","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=365;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-169aa4757c06fd71daf2aef3","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":158.86,"rawValue":"158.86","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=42;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-c900fda7a412557394e6f6ba","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":158.86,"rawValue":"158.86","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=525;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-b89d3fa8192e18087af82bea","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-09","metricId":"ECI Score","value":158.86,"rawValue":"158.86","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=355;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-d79b1475f73e11730760efe3","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-09-19","metricId":"ECI Score","value":144.55,"rawValue":"144.55","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=596;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-641ec2c1deb52fb813029e03","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-08","metricId":"ECI Score","value":153.9,"rawValue":"153.9","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=159;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-e01196b60c5ce6db58240df8","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-08","metricId":"ECI Score","value":153.9,"rawValue":"153.9","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=624;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-00b6c1c60ccaea9d25d59570","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-08","metricId":"ECI Score","value":153.9,"rawValue":"153.9","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=617;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-ecca46de11e5f0252d4ee605","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-08","metricId":"ECI Score","value":153.9,"rawValue":"153.9","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=472;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4b34b122f35ef8d139314632","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-12","metricId":"ECI Score","value":156.15,"rawValue":"156.15","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=7;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-b4572601a3ee4d0451de126f","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-12","metricId":"ECI Score","value":156.15,"rawValue":"156.15","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=354;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6461f542a66253abefd0817d","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_medium","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-12","metricId":"ECI Score","value":156.15,"rawValue":"156.15","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=350;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-a85506194a8d0611d1d30aac","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-12","metricId":"ECI Score","value":156.15,"rawValue":"156.15","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=523;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-a5d7811cd704ab0f593f9f91","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-12","metricId":"ECI Score","value":156.15,"rawValue":"156.15","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=4;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ebf8714929818e6edf2c1867","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-01-27","metricId":"ECI Score","value":148.22,"rawValue":"148.22","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=195;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1ddcfdb03070ca595f45993c","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-01-27","metricId":"ECI Score","value":148.22,"rawValue":"148.22","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=782;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-5506526224a559718503c25c","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-20","metricId":"ECI Score","value":151.0,"rawValue":"151.0","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=22;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-040172d8bd9b6c1a099371bd","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-12","metricId":"ECI Score","value":150.52,"rawValue":"150.52","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=50;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-691c088f8550632adb60a80c","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_high","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-16","metricId":"ECI Score","value":157.33,"rawValue":"157.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=40;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-03e21119daae018885dfdd41","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_low","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-16","metricId":"ECI Score","value":157.33,"rawValue":"157.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=39;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-1621c1361994c2d148eb70b9","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-16","metricId":"ECI Score","value":157.33,"rawValue":"157.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=126;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-461f4a160b7dd9584c5b1855","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-07-16","metricId":"ECI Score","value":157.33,"rawValue":"157.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=395;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e4686486e9d6bf89e429f5ee","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-25","metricId":"ECI Score","value":143.97,"rawValue":"143.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=78;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7d2440acd2fcb4964a0e52d6","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-25","metricId":"ECI Score","value":143.97,"rawValue":"143.97","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=121;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-acdfcdbbf6daa8bcd0bbaaae","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-16","metricId":"ECI Score","value":147.18,"rawValue":"147.18","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=62;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2bd0b8586949b16051d84929","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-02-16","metricId":"ECI Score","value":147.18,"rawValue":"147.18","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=119;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ed81d28be83f964d69f983d8","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-14","metricId":"ECI Score","value":145.03,"rawValue":"145.03","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=66;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b3a0dea0567d22487098c6d4","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-14","metricId":"ECI Score","value":145.03,"rawValue":"145.03","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=67;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d31bc8cc90c3fae43e5b7054","canonicalModelId":"qwen/qwen3.6-flash@2026-04-27","modelName":"Qwen3.6 Flash","modelRef":"qwen3.6-flash","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-27","metricId":"ECI Score","value":144.15,"rawValue":"144.15","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=68;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ddf9ab96735a708f416abfd6","canonicalModelId":"qwen/qwen3.6-flash@2026-04-27","modelName":"Qwen3.6 Flash","modelRef":"qwen3.6-flash_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-27","metricId":"ECI Score","value":144.15,"rawValue":"144.15","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=120;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-14199297801de9a20859a819","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-20","metricId":"ECI Score","value":149.33,"rawValue":"149.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=45;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-049a46ef675ebcd32c0917a9","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-04-20","metricId":"ECI Score","value":149.33,"rawValue":"149.33","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=114;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d7323e7a7eb59f773a7aa426","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-31","metricId":"ECI Score","value":149.37,"rawValue":"149.37","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=54;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7be294f4cd0f77a3f2986b4c","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-03-31","metricId":"ECI Score","value":149.37,"rawValue":"149.37","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=118;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-1fab0a2e28cf35eb1c5f93d3","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-05-19","metricId":"ECI Score","value":153.55,"rawValue":"153.55","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=47;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-690cc054303473f00222f8fa","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-02","metricId":"ECI Score","value":148.3,"rawValue":"148.3","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=70;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-15c5d6f4e950cdf0c7ddf417","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus_none","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-06-02","metricId":"ECI Score","value":148.3,"rawValue":"148.3","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=71;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-20861e8b2481167afed52280","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max_unknown","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-02","metricId":"ECI Score","value":156.42,"rawValue":"156.42","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=532;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e17aa49fd99483d38dc86464","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max_xhigh","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2026-08-02","metricId":"ECI Score","value":156.42,"rawValue":"156.42","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=124;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-08c906b79845e4725cdf59de","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/GLM-4.7","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-22","metricId":"ECI Score","value":143.94,"rawValue":"143.94","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=205;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-c2ed8bdc24d569240339ac31","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/glm-4.7","benchmarkId":"epoch-epoch_capabilities_index","benchmarkVersion":"2025-12-22","metricId":"ECI Score","value":143.94,"rawValue":"143.94","unit":"index","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=740;column=ECI Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-94fdc1cb2aa8198d70153e9f","canonicalModelId":"deepseek/v3.1@2025-08-21","modelName":"DeepSeek V3.1","modelRef":"DeepSeek-V3.1","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-08-21","metricId":"Overall score","value":58.0,"rawValue":"58.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=59;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-52544a2f15f1f2ed2a30fc94","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-06-01","metricId":"Overall score","value":60.4,"rawValue":"60.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=28;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-274db9774181bf8f8e888c06","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-10-15","metricId":"Overall score","value":61.4,"rawValue":"61.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=9;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3c25447be7c8211badc13526","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-05-22","metricId":"Overall score","value":61.1,"rawValue":"61.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=15;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c1c0e58c666de7cc65be3fa2","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-11-24","metricId":"Overall score","value":60.7,"rawValue":"60.7","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=21;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6af0334e2219f01c2613823a","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-02-05","metricId":"Overall score","value":60.0,"rawValue":"60.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=32;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c35e9a4305363fd3778a15ab","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-04-16","metricId":"Overall score","value":59.6,"rawValue":"59.6","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=40;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ef9e42333e59a2802469b296","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_24K","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-05-28","metricId":"Overall score","value":59.3,"rawValue":"59.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=44;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d2d7078eb2606bfc54618a2e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-05-28","metricId":"Overall score","value":58.5,"rawValue":"58.5","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=55;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-03d752f59cd304411bf41565","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-09-29","metricId":"Overall score","value":61.9,"rawValue":"61.9","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=4;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-339edc1e0c416fd53bd6d8e6","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_16K","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-02-17","metricId":"Overall score","value":61.2,"rawValue":"61.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=14;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-34dfc007df6cc3f9e3c4ffa7","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-02-17","metricId":"Overall score","value":59.6,"rawValue":"59.6","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=39;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-deb215924378147d76f25ee4","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_16K","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-06-30","metricId":"Overall score","value":59.9,"rawValue":"59.9","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=35;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-673f6e3b2036959ccce72bab","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-04-24","metricId":"Overall score","value":55.3,"rawValue":"55.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=75;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-def5b38df6ba8d344d2dfcf4","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-06-17","metricId":"Overall score","value":59.0,"rawValue":"59.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=50;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-cdb2d06f803af57d32f34e69","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-06-17","metricId":"Overall score","value":60.4,"rawValue":"60.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=26;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-73a48f30cd53e219b9a8adb2","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-02-19","metricId":"Overall score","value":59.0,"rawValue":"59.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=49;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1d21142b04c12c60174f289a","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_high","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-02-19","metricId":"Overall score","value":56.2,"rawValue":"56.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=74;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-4ea88e78d1851816f1f19f0a","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-05-19","metricId":"Overall score","value":59.2,"rawValue":"59.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=46;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d8af2b9dcc08b45fe4e1bb7d","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-02-11","metricId":"Overall score","value":61.0,"rawValue":"61.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=19;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1f91b63acd51e02289d90e91","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-08-07","metricId":"Overall score","value":61.4,"rawValue":"61.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=11;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-aee65c73e5d1532bcedbff22","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-08-07","metricId":"Overall score","value":61.0,"rawValue":"61.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=18;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1bd007720838575d9a59fc17","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-08-07","metricId":"Overall score","value":59.1,"rawValue":"59.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=47;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b2b141ec60532a2b53c8a2ab","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-11-13","metricId":"Overall score","value":58.1,"rawValue":"58.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=58;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-303e20d684601e8f17fda3fb","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-12-11","metricId":"Overall score","value":60.1,"rawValue":"60.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=31;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fc2764ff7b3044d647633bea","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-03-05","metricId":"Overall score","value":59.5,"rawValue":"59.5","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=41;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8fc84063e7cea42ae15db852","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-03-17","metricId":"Overall score","value":57.0,"rawValue":"57.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=70;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-107f30c05f50ec182b264628","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-03-17","metricId":"Overall score","value":57.3,"rawValue":"57.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=66;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4c479f2e794943f98dd6000e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2026-04-23","metricId":"Overall score","value":61.1,"rawValue":"61.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=16;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e75e23da558ad5ecc97ce465","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast","benchmarkId":"epoch-forecastbench_external","benchmarkVersion":"2025-09-19","metricId":"Overall score","value":60.5,"rawValue":"60.5","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=25;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-43971de7cf8aa3bb79158f16","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-06-01","metricId":"Main score","value":0.147,"rawValue":"0.147","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=27;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ede78b68e6d3986d64b2222f","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-06-09","metricId":"Main score","value":0.5348,"rawValue":"0.5348","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=2;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"claude-code","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a255168995e2978dd05f71ae","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-02-05","metricId":"Main score","value":0.2664,"rawValue":"0.2664","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=20;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"claude-code","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-38770ad6f0dbbe99265bc59a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-04-16","metricId":"Main score","value":0.3854,"rawValue":"0.3854","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=16;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"claude-code","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c4fd62a18227766be17a7231","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-05-28","metricId":"Main score","value":0.465,"rawValue":"0.465","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=6;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"claude-code","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dba0480e1e453076b2956421","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-07-24","metricId":"Main score","value":0.5338,"rawValue":"0.5338","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=3;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"claude-code","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-71446ab76bd9403b44e6c889","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-02-17","metricId":"Main score","value":0.24310000000000004,"rawValue":"0.24310000000000004","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=24;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"claude-code","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-13dfc6bd48bb1b61a053988c","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-06-30","metricId":"Main score","value":0.42730000000000007,"rawValue":"0.42730000000000007","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=10;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"claude-code","subject_type":"system"},"harnessId":"claude-code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-27bd4177d6211cd32e3fa5c2","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-07-31","metricId":"Main score","value":0.188,"rawValue":"0.188","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=25;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"chisel","subject_type":"system"},"harnessId":"chisel","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-75ddfafbe26112a6829acd3a","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_none","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-04-24","metricId":"Main score","value":0.1764,"rawValue":"0.1764","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=26;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-59efc1a08a0d92aa0bd378b9","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-07-21","metricId":"Main score","value":0.3437,"rawValue":"0.3437","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=17;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"chisel","subject_type":"system"},"harnessId":"chisel","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-17b0927f7185181fd2ae4fc3","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-08-13","metricId":"Main score","value":0.43590000000000007,"rawValue":"0.43590000000000007","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=8;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"chisel","subject_type":"system"},"harnessId":"chisel","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8cba59f43c139672c245517f","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_none","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-06-16","metricId":"Main score","value":0.245,"rawValue":"0.245","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=23;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4073bdbe6133faf507f4739e","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-03-17","metricId":"Main score","value":0.2704,"rawValue":"0.2704","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=19;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"codex","subject_type":"system"},"harnessId":"codex","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c5f0edbf4854db1b2de86f87","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-04-23","metricId":"Main score","value":0.42960000000000004,"rawValue":"0.42960000000000004","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=9;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"codex","subject_type":"system"},"harnessId":"codex","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b4fdf313cc2db63efb2f40bc","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-07-09","metricId":"Main score","value":0.3981,"rawValue":"0.3981","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=15;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"codex","subject_type":"system"},"harnessId":"codex","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0a5af6fe9473f47264a12443","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-07-09","metricId":"Main score","value":0.47490000000000004,"rawValue":"0.47490000000000004","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=5;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"codex","subject_type":"system"},"harnessId":"codex","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-019a1a70a84e06cedc9a7e0d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-07-09","metricId":"Main score","value":0.4131,"rawValue":"0.4131","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=14;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"codex","subject_type":"system"},"harnessId":"codex","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4f82871178a1d984bcbffdfc","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-07-08","metricId":"Main score","value":0.4244,"rawValue":"0.4244","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=11;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"grok-build","subject_type":"system"},"harnessId":"grok-build","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fed89667d781afe4c3e2a1f6","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-08-12","metricId":"Main score","value":0.4801000000000001,"rawValue":"0.4801000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=4;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"grok-build","subject_type":"system"},"harnessId":"grok-build","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0565b8b057a1a98cb94bf837","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-06-12","metricId":"Main score","value":0.3006,"rawValue":"0.3006","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=18;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-14a7818eee019bc453c24ea6","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_unknown","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-07-16","metricId":"Main score","value":0.44170000000000004,"rawValue":"0.44170000000000004","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=7;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1ece6d73862a47a13afb88d9","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"epoch-frontiercode_external","benchmarkVersion":"2026-06-02","metricId":"Main score","value":0.102,"rawValue":"0.102","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=31;column=Main score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"mini-swe-agent","subject_type":"system"},"harnessId":"mini-swe-agent","qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-bf9ec78ba72bb8b552484f94","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_32K","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-10-15","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":2.083333333333333,"rawValue":"0.020833333333333332","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=50;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b699fa684399df707bb9d620","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=64;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-23ea53b1e970a0fa3b929bd2","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_27K","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":4.166666666666666,"rawValue":"0.041666666666666664","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=68;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-50de2511ddcd92a0eac85221","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":4.166666666666666,"rawValue":"0.041666666666666664","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=41;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-32d74dc301d46379286616c4","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_16K","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":2.083333333333333,"rawValue":"0.020833333333333332","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=42;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-58d0571abb10d0dcfee32fdb","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_32K","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":4.166666666666666,"rawValue":"0.041666666666666664","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=43;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-aa444b58d4e89d2f4a51b9ac","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":14.583333333333334,"rawValue":"0.14583333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=27;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-5033eb55a0e1a1b047d9b49b","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_32K","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":20.833333333333336,"rawValue":"0.20833333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=26;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-f16e01092c306de50ba5d607","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_64K","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":20.833333333333336,"rawValue":"0.20833333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=25;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5283c6443ebc280abbc86acb","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":22.900000000000002,"rawValue":"0.229","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=24;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-86c1be7e4ca51b4e286567a3","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_xhigh","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":22.916666666666664,"rawValue":"0.22916666666666666","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=15;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fa721ddfff8c6b342d0958fb","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-05-28","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":31.25,"rawValue":"0.3125","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=2;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-857015e57c1d8b6e79d8891c","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":2.083333333333333,"rawValue":"0.020833333333333332","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=54;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5622074249600d7ecbc24d41","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":4.166666666666666,"rawValue":"0.041666666666666664","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=51;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-370a6ccbd3c3ff9c27a206e8","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_16K","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-02-17","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":8.3,"rawValue":"0.083","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=21;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f6e63e510b369e1f1d98c367","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"fireworks/deepseek-v3p2","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-12-01","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":2.1,"rawValue":"0.021","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=33;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-06596c038e135ec040b8a931","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"fireworks/kimi-k2p5","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-01-27","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":4.2,"rawValue":"0.042","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=28;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-176794f8d17194261f1db20e","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-06-17","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":4.166666666666666,"rawValue":"0.041666666666666664","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=31;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-510c760aaf4760741dfbf381","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-06-17","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":4.166666666666666,"rawValue":"0.041666666666666664","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=61;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2a8f3d25796be67ee19bb217","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-02-19","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":16.7,"rawValue":"0.167","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=22;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bc0a4ad75cd6e3481327121a","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":14.583333333333334,"rawValue":"0.14583333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=4;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-71725b2a2e43ef7847db3988","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-02-11","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":2.1,"rawValue":"0.021","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=23;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d1a3b8949416b84aa941838a","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-04-07","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":12.5,"rawValue":"0.125","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=9;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4fa5acde1ab818d29a2101c3","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":12.5,"rawValue":"0.125","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=49;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ab2542ffc397ee870f212d51","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":6.25,"rawValue":"0.0625","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=57;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b862580c9bf1e8a07ab517f9","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":6.25,"rawValue":"0.0625","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=47;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7be15e375bfa3163e1e98c09","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":4.166666666666666,"rawValue":"0.041666666666666664","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=58;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-86325ed9b5720e9450b7cc28","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=48;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4d53b9c22530963c33dbab43","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":2.083333333333333,"rawValue":"0.020833333333333332","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=59;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-7c7935c6a4970d82f9686a90","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":12.5,"rawValue":"0.125","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=46;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b84818dda52520d0c602d63e","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":4.166666666666666,"rawValue":"0.041666666666666664","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=45;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-eae00036a727ec55a1730f4e","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":18.8,"rawValue":"0.188","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=35;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-78484996daff0b388a5e0110","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_low","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":6.25,"rawValue":"0.0625","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=37;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-efe8e364ecf07029332eb18e","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":16.7,"rawValue":"0.167","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=36;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ff80999833d12d48f294b492","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":18.8,"rawValue":"0.188","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=34;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-056b72e9b6a36bb51337511c","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11-webapp","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":31.3,"rawValue":"0.313","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=30;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d7edcfdc4be5bb1d6c7a401b","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":27.1,"rawValue":"0.271","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=19;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8ff5ecf8fd3b27cddab80391","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":2.08,"rawValue":"0.0208","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=16;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-717db7da6dc77b912dd1abda","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":6.25,"rawValue":"0.0625","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=17;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1b42a7e920a5d178a194c4d1","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05-web-app","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":37.5,"rawValue":"0.375","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=20;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4c63c80020ca9f0a0de80dfa","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-04-20","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":14.580000000000002,"rawValue":"0.1458","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=11;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-374d24ef39057e16b72daa79","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-02-25","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=8;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0cee971842160c3829d424ce","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-02-16","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":2.083333333333333,"rawValue":"0.020833333333333332","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=5;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f8a9a027f13e5e5d70ad258c","canonicalModelId":"qwen/qwen3.6-flash@2026-04-27","modelName":"Qwen3.6 Flash","modelRef":"qwen3.6-flash","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-04-27","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=7;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-06fee06708b8cbac242878f1","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-04-20","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":4.166666666666666,"rawValue":"0.041666666666666664","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=3;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-97e2658a0b2ffeea60368c4b","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2026-03-31","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":8.333333333333332,"rawValue":"0.08333333333333333","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=6;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-de572d1b53f11e1ba91c90ce","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/GLM-4.7","benchmarkId":"epoch-frontiermath_tier_4","benchmarkName":"FrontierMath Tier 4 v2 (reported snapshot)","benchmarkVersion":"2025-12-22","benchmarkVersionHint":"frontiermath@v2-tier-4","metricId":"mean_score","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=29;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-416d843faa010022166c3c63","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_unknown","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-06-09","metricId":"Implementation rank","value":2.0,"rawValue":"2.0","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=9;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9c901a6c379d755355d96055","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-02-05","metricId":"Implementation rank","value":9.2,"rawValue":"9.2","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=4;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9b34a32fc0faef6e1553c176","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-04-16","metricId":"Implementation rank","value":7.2,"rawValue":"7.2","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=3;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cf972abb1ff992ac30cf8b2b","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_unknown","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-05-28","metricId":"Implementation rank","value":5.4,"rawValue":"5.4","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=10;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-994e05ac310588b5ba638f7e","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_unknown","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-04-24","metricId":"Implementation rank","value":13.0,"rawValue":"13.0","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=12;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-85e3190298991ebc9cd53745","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-02-19","metricId":"Implementation rank","value":13.8,"rawValue":"13.8","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=6;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Gemini CLI","subject_type":"system"},"harnessId":"Gemini CLI","qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-885648dc9f5a2a1148046722","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-04-07","metricId":"Implementation rank","value":12.9,"rawValue":"12.9","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=11;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-feb6a2103f712176e44dc37a","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_unknown","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-06-16","metricId":"Implementation rank","value":6.3,"rawValue":"6.3","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=15;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4ae7fa2f783866dc0c374096","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3_unknown","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-08-14","metricId":"Implementation rank","value":4.0,"rawValue":"4.0","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=18;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-adb74b255bba76a9b863cb66","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_unknown","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-03-05","metricId":"Implementation rank","value":8.7,"rawValue":"8.7","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=5;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex","subject_type":"system"},"harnessId":"Codex","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d2d54b0d15331230f8451870","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-04-23","metricId":"Implementation rank","value":8.8,"rawValue":"8.8","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=2;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex","subject_type":"system"},"harnessId":"Codex","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-aa8dfebcc89f0af203caeb87","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_unknown","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-07-08","metricId":"Implementation rank","value":6.2,"rawValue":"6.2","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=14;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Grok CLI","subject_type":"system"},"harnessId":"Grok CLI","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-26c88ce32ae61bb135be2db0","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_unknown","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-08-12","metricId":"Implementation rank","value":3.3,"rawValue":"3.3","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=17;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Grok CLI","subject_type":"system"},"harnessId":"Grok CLI","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-954e13ab2034ecb61db3eade","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-01-27","metricId":"Implementation rank","value":14.5,"rawValue":"14.5","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=8;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Kimi CLI","subject_type":"system"},"harnessId":"Kimi CLI","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-42feab8cc12bf7b0a38d35e4","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-04-20","metricId":"Implementation rank","value":11.7,"rawValue":"11.7","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=7;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Kimi CLI","subject_type":"system"},"harnessId":"Kimi CLI","qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c2cd58f13c57b8e6037f0f75","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"epoch-frontierswe_external","benchmarkVersion":"2026-03-31","metricId":"Implementation rank","value":15.8,"rawValue":"15.8","unit":"rank","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontierswe_external.csv:row=13;column=Implementation rank","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Qwen Code","subject_type":"system"},"harnessId":"Qwen Code","qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a4093c93b7c9955b8902ec57","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMax-M2.7","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-03-18","metricId":"Overall score","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=23;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-17120088921f7ff19a25db88","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-06-01","metricId":"Overall score","value":0.9136320401503086,"rawValue":"0.009136320401503086","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=17;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-328dc4e58e97d74acb7ec1d7","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-06-09","metricId":"Overall score","value":74.47198828312318,"rawValue":"0.7447198828312318","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=3;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f1ee0cca3575f6e05d54cb2d","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-02-05","metricId":"Overall score","value":44.11559216788414,"rawValue":"0.4411559216788414","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=12;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-092eabfcf579ee6005e1c91b","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-04-16","metricId":"Overall score","value":43.8113618760859,"rawValue":"0.43811361876085897","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=13;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e163c80d65a4a287bf541a5a","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-05-28","metricId":"Overall score","value":70.90450062635527,"rawValue":"0.7090450062635527","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=4;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-32839b702caf926f78a27060","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-07-24","metricId":"Overall score","value":79.59852171423216,"rawValue":"0.7959852171423216","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=2;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5263608d9e3a0cda8e267d35","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-02-17","metricId":"Overall score","value":48.756634832044746,"rawValue":"0.4875663483204475","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=9;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-88f1fa6ead8cc2dce32241ef","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-06-30","metricId":"Overall score","value":65.26299757455419,"rawValue":"0.6526299757455419","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=6;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c0889ed8dc2ff63004b8ddef","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-02-19","metricId":"Overall score","value":0.8462790979968672,"rawValue":"0.008462790979968672","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=20;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-33f2e9218b439dca5a81895f","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-05-19","metricId":"Overall score","value":6.701366402557545,"rawValue":"0.06701366402557545","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=15;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6f23e45c240d48f41019108f","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-04-07","metricId":"Overall score","value":0.0,"rawValue":"0.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=24;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4bdae6c21934cd1fcfe2ea5e","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-06-16","metricId":"Overall score","value":0.017218231881747016,"rawValue":"0.00017218231881747017","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=22;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ac852cd6115d100729a67455","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-03-05","metricId":"Overall score","value":45.14817826662626,"rawValue":"0.4514817826662626","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=11;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8d08a25db7250e97db320d3d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-04-23","metricId":"Overall score","value":53.22037178151383,"rawValue":"0.5322037178151383","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=7;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a09301f76a459c867e65f984","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-07-09","metricId":"Overall score","value":52.59227864005247,"rawValue":"0.5259227864005247","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=8;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5ab062fe6ce8a2a400fb7430","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-07-08","metricId":"Overall score","value":65.4126756585321,"rawValue":"0.6541267565853209","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=5;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-67fc7118b9dfb22bfe985f84","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-04-20","metricId":"Overall score","value":0.8624932115315737,"rawValue":"0.008624932115315736","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=18;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-cb0b5fed1b936089a4655db9","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-06-12","metricId":"Overall score","value":0.8526563108968812,"rawValue":"0.008526563108968812","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=19;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c6f49418057636b26f67a69d","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_unknown","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-07-16","metricId":"Overall score","value":48.346855379537594,"rawValue":"0.48346855379537595","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=10;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ba7cc002c9e522b00d5f164a","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"epoch-gbaeval_external","benchmarkVersion":"2026-05-19","metricId":"Overall score","value":0.40666088933695027,"rawValue":"0.004066608893369503","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gbaeval_external.csv:row=21;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-11dcce2aa0c91b199af89013","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-06-09","metricId":"GDP.pdf score","value":0.298,"rawValue":"0.298","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=15;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-03342d53cf4475446b59c399","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_unknown","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-06-09","metricId":"GDP.pdf score","value":0.3,"rawValue":"0.3","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=2;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"06/06/2026","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://surgehq.ai/leaderboards/gdp-pdf"},{"id":"pub-492423a37a5eb96bf1cdfde1","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-04-16","metricId":"GDP.pdf score","value":0.21,"rawValue":"0.21","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=5;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"06/06/2026","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-f02b3aac76fe2b72746b3609","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-05-28","metricId":"GDP.pdf score","value":0.24,"rawValue":"0.24","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=4;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"06/06/2026","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-4f3b659462161f9fcaa5e413","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-07-24","metricId":"GDP.pdf score","value":0.24,"rawValue":"0.24","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=17;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-aa33769524c8e2111e8665f5","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-02-17","metricId":"GDP.pdf score","value":0.18,"rawValue":"0.18","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=20;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-938b3c722a1a14e5d49a0c29","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-02-19","metricId":"GDP.pdf score","value":0.17,"rawValue":"0.17","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=6;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"06/06/2026","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-2422d85f6774468b02890de5","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_high","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-07-21","metricId":"GDP.pdf score","value":0.1,"rawValue":"0.1","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=24;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-a2fe1709d8f1275c321e9986","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_unknown","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-05-19","metricId":"GDP.pdf score","value":0.14,"rawValue":"0.14","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=7;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"06/06/2026","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-856514968011f1e8510bb2a2","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-07-21","metricId":"GDP.pdf score","value":0.14,"rawValue":"0.14","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=27;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-f9d3017f89c8fcd88fcea459","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_unknown","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-07-21","metricId":"GDP.pdf score","value":0.14,"rawValue":"0.14","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=22;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-7e89aebb49c5f98c0d599577","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-04-23","metricId":"GDP.pdf score","value":0.26,"rawValue":"0.26","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=3;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"06/06/2026","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-be2d8d70cf988dadeaf4aaec","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_unknown","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-07-09","metricId":"GDP.pdf score","value":0.227,"rawValue":"0.227","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=18;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-1673318876635b60705e79c8","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-07-09","metricId":"GDP.pdf score","value":0.307,"rawValue":"0.307","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=26;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-503fe2d72e97ff575b31b407","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_unknown","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-07-09","metricId":"GDP.pdf score","value":0.307,"rawValue":"0.307","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=14;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-6c218caef7dbb933fd2724e4","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_unknown","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-07-09","metricId":"GDP.pdf score","value":0.247,"rawValue":"0.247","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=16;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-fe3e9d021318b56ee0321318","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-07-08","metricId":"GDP.pdf score","value":0.14,"rawValue":"0.14","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=23;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-a900f736d40a157798ed4164","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-04-20","metricId":"GDP.pdf score","value":0.12,"rawValue":"0.12","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=8;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"06/06/2026","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-af83131a9b6b752591fbdb0f","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-gdp_pdf_external","benchmarkVersion":"2026-07-16","metricId":"GDP.pdf score","value":0.19,"rawValue":"0.19","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdp_pdf_external.csv:row=19;column=GDP.pdf score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf"},{"id":"pub-c001a9fd7505d46d0d68fdee","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-gdpval_external","benchmarkVersion":"2025-11-24","metricId":"Win Rate (%)","value":45.5,"rawValue":"0.455","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdpval_external.csv:row=3;column=Win Rate (%)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d9f6a6d64f8f3d65ebb6ca25","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-gdpval_external","benchmarkVersion":"2025-09-29","metricId":"Win Rate (%)","value":42.5,"rawValue":"0.425","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdpval_external.csv:row=5;column=Win Rate (%)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5526125b30708640b9afabc3","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-gdpval_external","benchmarkVersion":"2025-06-17","metricId":"Win Rate (%)","value":23.3,"rawValue":"0.233","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdpval_external.csv:row=10;column=Win Rate (%)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-738a687ef467a99fcaaf04b2","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-gdpval_external","benchmarkVersion":"2025-08-07","metricId":"Win Rate (%)","value":34.8,"rawValue":"0.348","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdpval_external.csv:row=7;column=Win Rate (%)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4e814c902891ef5e80452d26","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_none","benchmarkId":"epoch-gdpval_external","benchmarkVersion":"2025-12-11","metricId":"Win Rate (%)","value":49.7,"rawValue":"0.49700000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gdpval_external.csv:row=2;column=Win Rate (%)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e84ef536963e78b83526b6bb","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_32K","benchmarkId":"epoch-geobench_external","benchmarkVersion":"2025-05-22","metricId":"ACW Avg Score","value":2915.0,"rawValue":"2915","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"geobench_external.csv:row=27;column=ACW Avg Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://geobench.org/"},{"id":"pub-fc64486fbcf4cd2e916a61a0","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-geobench_external","benchmarkVersion":"2025-11-24","metricId":"ACW Avg Score","value":3877.0,"rawValue":"3877","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"geobench_external.csv:row=12;column=ACW Avg Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5ad749005c6c2cc24ac83b17","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-geobench_external","benchmarkVersion":"2025-08-07","metricId":"ACW Avg Score","value":3937.0,"rawValue":"3937","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"geobench_external.csv:row=7;column=ACW Avg Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://geobench.org/"},{"id":"pub-64f98ddf3b0e6cf2567ff52d","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-07-31","metricId":"Score OPT@1","value":0.049,"rawValue":"0.049","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=28;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://gso-bench.github.io/leaderboard.html"},{"id":"pub-5095df32b7920a77cd241941","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-05-22","metricId":"Score OPT@1","value":0.069,"rawValue":"0.069","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=23;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://gso-bench.github.io/leaderboard.html"},{"id":"pub-7085d52fb24367d4a4094a66","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_unknown","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-05-22","metricId":"Score OPT@1","value":0.06860000000000001,"rawValue":"0.06860000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=25;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-64f54d890dd628ffeb8c7c1a","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-11-24","metricId":"Score OPT@1","value":0.265,"rawValue":"0.265","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=12;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://gso-bench.github.io/leaderboard.html"},{"id":"pub-d6322ee69b61180c3062a1ef","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-11-24","metricId":"Score OPT@1","value":0.2647,"rawValue":"0.2647","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=13;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5703d23524201c086c76c6d5","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"epoch-gso_external","benchmarkVersion":"2026-02-05","metricId":"Score OPT@1","value":0.3333,"rawValue":"0.3333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=9;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-431481067d67f28c9ed7bc28","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_high","benchmarkId":"epoch-gso_external","benchmarkVersion":"2026-02-05","metricId":"Score OPT@1","value":0.412,"rawValue":"0.412","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=5;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-5ec0052282fe352bc9fbce0c","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-gso_external","benchmarkVersion":"2026-02-05","metricId":"Score OPT@1","value":0.3333,"rawValue":"0.3333","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=8;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-30977a414dc3b419bf7b4f64","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"epoch-gso_external","benchmarkVersion":"2026-04-16","metricId":"Score OPT@1","value":0.4412,"rawValue":"0.4412","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=4;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-4a349aebcd7a7ab5ce37aebd","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_high","benchmarkId":"epoch-gso_external","benchmarkVersion":"2026-04-16","metricId":"Score OPT@1","value":0.4412,"rawValue":"0.4412","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=3;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8830684e2bab63dfba159a0a","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_unknown","benchmarkId":"epoch-gso_external","benchmarkVersion":"2026-05-28","metricId":"Score OPT@1","value":0.4706,"rawValue":"0.4706","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=2;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-344acec0bcf169072f1ba4a7","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-09-29","metricId":"Score OPT@1","value":0.147,"rawValue":"0.147","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=18;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://gso-bench.github.io/leaderboard.html"},{"id":"pub-97eae7feac8fd257b820e1cd","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-09-29","metricId":"Score OPT@1","value":0.1471,"rawValue":"0.1471","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=17;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fe4c6ea0691f0b311593ba60","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_unknown","benchmarkId":"epoch-gso_external","benchmarkVersion":"2026-06-30","metricId":"Score OPT@1","value":0.3725,"rawValue":"0.3725","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=7;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b45188cb124c80071bc36559","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-06-17","metricId":"Score OPT@1","value":0.0392,"rawValue":"0.0392","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=31;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://gso-bench.github.io/leaderboard.html"},{"id":"pub-c6b176bad48dbb4de2c20b11","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-gso_external","benchmarkVersion":"2026-02-19","metricId":"Score OPT@1","value":0.2255,"rawValue":"0.2255","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=15;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ca0ef35faaf4557804e3b5a3","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-08-07","metricId":"Score OPT@1","value":0.06860000000000001,"rawValue":"0.06860000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=24;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://gso-bench.github.io/leaderboard.html"},{"id":"pub-5ee75eae96b0dbb2546679fe","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-11-13","metricId":"Score OPT@1","value":0.137,"rawValue":"0.137","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=20;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-5f1e833895d996b6a1a3ea37","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-11-13","metricId":"Score OPT@1","value":0.1373,"rawValue":"0.1373","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=19;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-200748eb7353dcf3ade4ff26","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-gso_external","benchmarkVersion":"2025-12-11","metricId":"Score OPT@1","value":0.2745,"rawValue":"0.2745","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=11;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9729a15f8d3524da16bc2676","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"epoch-gso_external","benchmarkVersion":"2026-03-05","metricId":"Score OPT@1","value":0.2549,"rawValue":"0.2549","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=14;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c9c41e4d3e0e36aa94a90295","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-gso_external","benchmarkVersion":"2026-03-05","metricId":"Score OPT@1","value":0.31370000000000003,"rawValue":"0.31370000000000003","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=10;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-da7245d7fe9c8ca6449fa4c0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-gso_external","benchmarkVersion":"2026-04-23","metricId":"Score OPT@1","value":0.402,"rawValue":"0.402","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gso_external.csv:row=6;column=Score OPT@1","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-98e34c60f6c725281b1f8c9f","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"epoch-lech_mazur_writing_external","benchmarkVersion":"2025-05-28","metricId":"Mean score","value":8.19,"rawValue":"8.19","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"lech_mazur_writing_external.csv:row=46;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://github.com/lechmazur/Writing"},{"id":"pub-6f8af0565b66df237eb9e378","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"epoch-lech_mazur_writing_external","benchmarkVersion":"2025-05-22","metricId":"Mean score","value":8.31,"rawValue":"8.31","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"lech_mazur_writing_external.csv:row=44;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://github.com/lechmazur/Writing"},{"id":"pub-e8be7808dd51ed3e3bc4682c","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_16K","benchmarkId":"epoch-lech_mazur_writing_external","benchmarkVersion":"2025-05-22","metricId":"Mean score","value":8.36,"rawValue":"8.36","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"lech_mazur_writing_external.csv:row=43;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://github.com/lechmazur/Writing"},{"id":"pub-e30dde770f48a02064efdac3","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-lech_mazur_writing_external","benchmarkVersion":"2025-06-17","metricId":"Mean score","value":8.38,"rawValue":"8.38","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"lech_mazur_writing_external.csv:row=36;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://github.com/lechmazur/Writing"},{"id":"pub-d6e489fbd6e1ab133864c3cc","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-lech_mazur_writing_external","benchmarkVersion":"2025-08-07","metricId":"Mean score","value":8.6,"rawValue":"8.6","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"lech_mazur_writing_external.csv:row=33;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://github.com/lechmazur/Writing"},{"id":"pub-9547c487a1c20af9976da4a0","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-lech_mazur_writing_external","benchmarkVersion":"2025-08-07","metricId":"Mean score","value":8.31,"rawValue":"8.31","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"lech_mazur_writing_external.csv:row=45;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://github.com/lechmazur/Writing"},{"id":"pub-31e87cf3be22ae4481d00a99","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"epoch-math_level_5","benchmarkVersion":"2025-05-28","metricId":"mean_score","value":96.63897280966768,"rawValue":"0.9663897280966768","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=13;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-714e676ef232612ceaa029e4","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-math_level_5","benchmarkVersion":"2025-10-15","metricId":"mean_score","value":86.91465256797582,"rawValue":"0.8691465256797583","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=6;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1cb2d090bee9b5b0dd9e1e0c","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_32K","benchmarkId":"epoch-math_level_5","benchmarkVersion":"2025-10-15","metricId":"mean_score","value":96.36087945413192,"rawValue":"0.9636087945413192","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=4;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-381f081247556ee0b7ed4b09","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"epoch-math_level_5","benchmarkVersion":"2025-05-22","metricId":"mean_score","value":85.04531722054381,"rawValue":"0.850453172205438","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=15;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a6d384de420f1ae339e18343","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"epoch-math_level_5","benchmarkVersion":"2025-09-29","metricId":"mean_score","value":97.73413897280967,"rawValue":"0.9773413897280967","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=5;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b3486502e42c67b6c9708e19","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-math_level_5","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":98.13066465256797,"rawValue":"0.9813066465256798","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=3;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-c450de1365772e2fed9da768","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-math_level_5","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":97.92296072507553,"rawValue":"0.9792296072507553","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=8;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ed5d0e6d09c77682462a5882","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-math_level_5","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":97.8474320241692,"rawValue":"0.9784743202416919","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=2;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b9b2f0331b3dd8fe465435f4","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-math_level_5","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":96.79003021148036,"rawValue":"0.9679003021148036","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=11;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-8c9c06b3f0e806c925cf7252","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"epoch-math_level_5","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":94.90181268882175,"rawValue":"0.9490181268882175","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=10;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c23e79fd1349eb586df8b47b","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"epoch-math_level_5","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":95.2416918429003,"rawValue":"0.952416918429003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=9;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f7630393b6ec5995a7cb89f4","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2025-05-28","metricId":"average_score","value":0.537826,"rawValue":"0.537826","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=31;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/"},{"id":"pub-02302de729d2b6b717b24d9a","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_16K","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2025-05-22","metricId":"average_score","value":0.639292,"rawValue":"0.639292","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=22;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/"},{"id":"pub-42fbe2db4fe495fd38df20d5","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_unknown","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2025-05-22","metricId":"average_score","value":0.615452,"rawValue":"0.615452","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=20;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8d26614c69e8e5949634a028","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2025-11-24","metricId":"average_score","value":0.73007,"rawValue":"0.73007","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=9;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f7d9b45e0427b8dab9ecb4c9","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_16K","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2025-11-24","metricId":"average_score","value":0.74969,"rawValue":"0.74969","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=10;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/"},{"id":"pub-4c8e5c398ec66bef427cff81","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2026-02-05","metricId":"average_score","value":0.788644,"rawValue":"0.788644","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=3;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b01cfba365add511e723a1ae","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_16K","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2025-09-29","metricId":"average_score","value":0.673842,"rawValue":"0.673842","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=15;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/"},{"id":"pub-3fd099beb13d63e1c2a2557c","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2026-02-19","metricId":"average_score","value":0.77035,"rawValue":"0.77035","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=4;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f15eeb8a79cae7a32d282357","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2025-08-07","metricId":"average_score","value":0.693697,"rawValue":"0.693697","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=13;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3a168a5476f62cfa143e7b55","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2025-08-07","metricId":"average_score","value":0.696053,"rawValue":"0.696053","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=14;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/"},{"id":"pub-26f92cf67e0c3475d3ca4af1","canonicalModelId":"openai/gpt-5.1-codex-max@2025-11-19","modelName":"GPT-5.1 Codex Max","modelRef":"gpt-5.1-codex-max","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2025-11-19","metricId":"average_score","value":0.708301,"rawValue":"0.708301","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=12;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/"},{"id":"pub-5ff58d445dd61b2eb6401e49","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2025-12-11","metricId":"average_score","value":0.752875,"rawValue":"0.752875","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=5;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4d1373a61ca0a5d96e1658ad","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2026-02-05","metricId":"average_score","value":0.745439,"rawValue":"0.745439","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=6;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f68c33cee2761d974f02a9a9","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_unknown","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2026-03-05","metricId":"average_score","value":0.743443,"rawValue":"0.743443","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=7;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-49112c8e4356b027a0e9811d","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-metr_time_horizons_external","benchmarkVersion":"2026-03-05","metricId":"average_score","value":0.743443,"rawValue":"0.743443","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=8;column=average_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f1cde3caef874d8a450c855b","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-mirrorcode","benchmarkVersion":"2026-06-09","metricId":"mean_score","value":63.88888888888889,"rawValue":"0.638888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mirrorcode.csv:row=7;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a015166c17a7d8eadb61e814","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_high","benchmarkId":"epoch-mirrorcode","benchmarkVersion":"2026-04-16","metricId":"mean_score","value":31.11111111111111,"rawValue":"0.3111111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mirrorcode.csv:row=2;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ac1f261f9f047aa68789c2e7","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_high","benchmarkId":"epoch-mirrorcode","benchmarkVersion":"2026-02-19","metricId":"mean_score","value":8.888888888888888,"rawValue":"0.08888888888888888","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mirrorcode.csv:row=3;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3d2c84e5299a0beb9b186e4d","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"epoch-mirrorcode","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":15.555555555555555,"rawValue":"0.15555555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mirrorcode.csv:row=5;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a170508a39f0c79e6a0553fd","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-mirrorcode","benchmarkVersion":"2026-04-23","metricId":"mean_score","value":10.0,"rawValue":"0.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mirrorcode.csv:row=4;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9d8a8406a16d629cb107db08","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_high","benchmarkId":"epoch-mirrorcode","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":20.0,"rawValue":"0.2","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mirrorcode.csv:row=6;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b8871333dc7d4a7295b835ed","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-06-09","metricId":"mean_score","value":52.0,"rawValue":"0.52","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=35;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1f828e21606f96aff62b2a76","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_48K","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2025-11-24","metricId":"mean_score","value":22.0,"rawValue":"0.22","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=48;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ede51262c9797192b0e67c91","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-02-05","metricId":"mean_score","value":15.0,"rawValue":"0.15","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=11;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0c84c680979551c815ef803b","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_low","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-02-05","metricId":"mean_score","value":7.000000000000001,"rawValue":"0.07","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=12;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-896043a25b526b0b8ce8aa88","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-02-05","metricId":"mean_score","value":25.0,"rawValue":"0.25","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=51;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f8d5018ec2fbe8724c71e376","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-04-16","metricId":"mean_score","value":13.0,"rawValue":"0.13","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=10;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c34073b8c1b7fcb8720ab90a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-04-16","metricId":"mean_score","value":28.000000000000004,"rawValue":"0.28","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=52;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-48505c858cce5b488f432e6a","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-05-28","metricId":"mean_score","value":36.0,"rawValue":"0.36","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=49;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-351dcb8297fe8c4d0f69732d","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_xhigh","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-05-28","metricId":"mean_score","value":31.0,"rawValue":"0.31","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=45;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d24321639de37ab4c4557d15","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-24","metricId":"mean_score","value":37.0,"rawValue":"0.37","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=9;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-e520c159ca0ab76a888a61c2","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-24","metricId":"mean_score","value":59.0,"rawValue":"0.59","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=50;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a0814b6324eef39889db2585","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_48K","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2025-09-29","metricId":"mean_score","value":17.0,"rawValue":"0.17","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=46;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e3599a33d7a37a408e3d3fa6","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-02-17","metricId":"mean_score","value":14.000000000000002,"rawValue":"0.14","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=7;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-326d1c13001d152a8d8ef59b","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_low","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-02-17","metricId":"mean_score","value":16.0,"rawValue":"0.16","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=8;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-0c9ee482ea6bffd2a5afd405","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-06-30","metricId":"mean_score","value":16.0,"rawValue":"0.16","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=16;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-e09fbdea4824f9dcd784bded","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-06-30","metricId":"mean_score","value":35.0,"rawValue":"0.35","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=37;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3bd9e637e80089275248478c","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-31","metricId":"mean_score","value":34.0,"rawValue":"0.34","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=26;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8e2ebd31719c47011aa49dbe","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-08-13","metricId":"mean_score","value":43.0,"rawValue":"0.43","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=2;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ea3df8f1b2087e83f030a800","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_high","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-02-19","metricId":"mean_score","value":34.0,"rawValue":"0.34","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=43;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1e2458022262a9398b21ea1b","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_low","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-02-19","metricId":"mean_score","value":28.999999999999996,"rawValue":"0.29","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=15;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-7dca6686ab3458c80dfcbbfe","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_medium","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-02-19","metricId":"mean_score","value":32.0,"rawValue":"0.32","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=14;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-3eb3b2f8563033aa80590476","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_low","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":19.0,"rawValue":"0.19","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=25;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b49a909f2cfa7e61eee9f344","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_minimal","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":12.0,"rawValue":"0.12","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=24;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6c15df14f27d0e2a92468d70","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-05-19","metricId":"mean_score","value":32.0,"rawValue":"0.32","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=44;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-cd6ca5e6f952ece3621361e1","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_low","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-05-19","metricId":"mean_score","value":28.000000000000004,"rawValue":"0.28","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=18;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-90714972ed52eef7eca6dfc2","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":30.0,"rawValue":"0.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=33;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b3923c9fca29c0ee9d78ab40","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_low","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":23.0,"rawValue":"0.23","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=20;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4ec88da5986f25abf8570104","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_minimal","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":25.0,"rawValue":"0.25","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=19;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-bac7e4505c81430925ddb48c","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-08-13","metricId":"mean_score","value":37.0,"rawValue":"0.37","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=4;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dcef9d33dcfa98e367b23b70","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":23.0,"rawValue":"0.23","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=32;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-af807cba791b4a93350164ff","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2025-12-11","metricId":"mean_score","value":23.0,"rawValue":"0.23","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=6;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fbb73d158f32c5001c24494f","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":37.0,"rawValue":"0.37","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=53;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e33ecb83284eb9559a069d0e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-04-23","metricId":"mean_score","value":52.0,"rawValue":"0.52","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=42;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6b9037508832c46568dd0b4c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_none","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-04-23","metricId":"mean_score","value":32.0,"rawValue":"0.32","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=13;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ce140fb69199623be0ddf816","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-04-23","metricId":"mean_score","value":56.00000000000001,"rawValue":"0.56","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=54;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b7eecfacfafcf204d84270d8","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":21.0,"rawValue":"0.21","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=38;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a5403f685a01369cbb507272","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":57.99999999999999,"rawValue":"0.58","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=40;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-537c35b8f6eaca0ba92a350b","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":35.0,"rawValue":"0.35","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=39;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-dfbe040e14ec9829b898af1e","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_xhigh","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-08-12","metricId":"mean_score","value":34.0,"rawValue":"0.34","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=5;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3c95a8e44cb7d3da0d28bf7b","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-04-20","metricId":"mean_score","value":18.0,"rawValue":"0.18","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=55;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8efe0299b70261590cfb93f2","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-07-16","metricId":"mean_score","value":26.0,"rawValue":"0.26","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=36;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ac3c8f604eea00260ce7f48f","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash_none","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-02-25","metricId":"mean_score","value":15.0,"rawValue":"0.15","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=31;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4cad57f18542e9d96cbfbca2","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-02-16","metricId":"mean_score","value":16.0,"rawValue":"0.16","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=29;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-77382043352da4e045d8f9a0","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus_none","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-02-16","metricId":"mean_score","value":15.0,"rawValue":"0.15","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=28;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-67ceaf93c2ea74bed7bb53dd","canonicalModelId":"qwen/qwen3.6-flash@2026-04-27","modelName":"Qwen3.6 Flash","modelRef":"qwen3.6-flash_none","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-04-27","metricId":"mean_score","value":14.000000000000002,"rawValue":"0.14","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=30;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-164503134b5941db3ea6337e","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview_none","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-04-20","metricId":"mean_score","value":16.0,"rawValue":"0.16","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=17;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6625fd53a7ab7d53b3041f6d","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus_none","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-03-31","metricId":"mean_score","value":27.0,"rawValue":"0.27","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=27;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-106b9740e89c59e358b46de1","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-05-19","metricId":"mean_score","value":32.0,"rawValue":"0.32","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=41;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4f0497bde3740ca9b4b113a5","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max_xhigh","benchmarkId":"epoch-mystery_game_puzzles","benchmarkVersion":"2026-08-02","metricId":"mean_score","value":38.0,"rawValue":"0.38","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mystery_game_puzzles.csv:row=34;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a7f0f685eb60f5059f5c6d48","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-osworld_2_external","benchmarkVersion":"2026-06-01","metricId":"Binary accuracy","value":4.6,"rawValue":"0.046","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"osworld_2_external.csv:row=9;column=Binary accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9df4b62d2ceee7eeccaae767","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-osworld_2_external","benchmarkVersion":"2026-04-16","metricId":"Binary accuracy","value":18.2,"rawValue":"0.182","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"osworld_2_external.csv:row=4;column=Binary accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9091f9419adb3fd6e655f55d","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-osworld_2_external","benchmarkVersion":"2026-04-16","metricId":"Binary accuracy","value":13.900000000000002,"rawValue":"0.139","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"osworld_2_external.csv:row=5;column=Binary accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-cd9f7c4dfcdfa7b902ce664e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-osworld_2_external","benchmarkVersion":"2026-05-28","metricId":"Binary accuracy","value":18.52,"rawValue":"0.1852","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"osworld_2_external.csv:row=3;column=Binary accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-acc178e1c0386dd7425af869","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-osworld_2_external","benchmarkVersion":"2026-05-28","metricId":"Binary accuracy","value":20.6,"rawValue":"0.20600000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"osworld_2_external.csv:row=2;column=Binary accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-a3a4bcc00a53603e20b20d6b","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"epoch-osworld_2_external","benchmarkVersion":"2026-02-17","metricId":"Binary accuracy","value":8.3,"rawValue":"0.083","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"osworld_2_external.csv:row=7;column=Binary accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b3adc8bbed0674f31868a091","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_medium","benchmarkId":"epoch-osworld_2_external","benchmarkVersion":"2026-02-17","metricId":"Binary accuracy","value":9.3,"rawValue":"0.09300000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"osworld_2_external.csv:row=6;column=Binary accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9c0bb7ad43b7dec7fdaeac72","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-osworld_2_external","benchmarkVersion":"2026-04-23","metricId":"Binary accuracy","value":13.0,"rawValue":"0.13","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"osworld_2_external.csv:row=8;column=Binary accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a49b3286d650378a4540b7de","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-osworld_2_external","benchmarkVersion":"2026-04-20","metricId":"Binary accuracy","value":4.6,"rawValue":"0.046","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"osworld_2_external.csv:row=10;column=Binary accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8daac7992c796157877bceb4","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"epoch-osworld_2_external","benchmarkVersion":"2026-06-02","metricId":"Binary accuracy","value":2.8,"rawValue":"0.027999999999999997","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"osworld_2_external.csv:row=11;column=Binary accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-cbe8150ac2226bd252a40115","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-05-28","metricId":"mean_score","value":66.38888888888889,"rawValue":"0.6638888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=161;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-904bc75de143ad60661c1376","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-06-01","metricId":"mean_score","value":71.11111111111111,"rawValue":"0.7111111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=8;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-93c45c10c3a808f729ab87c5","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-06-09","metricId":"mean_score","value":100.0,"rawValue":"1.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=72;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d2d7d20f6b1495b0e41994c2","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-06-09","metricId":"mean_score","value":97.77777777777777,"rawValue":"0.9777777777777777","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=71;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-920a584740d5001f4a34984b","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-06-09","metricId":"mean_score","value":99.72222222222223,"rawValue":"0.9972222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=107;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-3e7b3a79713860f819342b9c","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-10-15","metricId":"mean_score","value":35.833333333333336,"rawValue":"0.35833333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=151;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-7edaa1609344e61db02ac356","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_32K","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-10-15","metricId":"mean_score","value":66.66666666666666,"rawValue":"0.6666666666666666","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=149;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-05d1473674b8c4ccb5518803","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-05-22","metricId":"mean_score","value":42.22222222222222,"rawValue":"0.4222222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=167;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-d19c7c3a2dd1d1414d283395","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_16K","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-05-22","metricId":"mean_score","value":60.0,"rawValue":"0.6","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=165;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c1457b2a5308f0b087a641be","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_27K","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-05-22","metricId":"mean_score","value":64.44444444444444,"rawValue":"0.6444444444444445","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=162;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-60f238687cdb1a25c471409d","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-11-24","metricId":"mean_score","value":48.05555555555556,"rawValue":"0.48055555555555557","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=134;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ae050b154b7655b1d02a7737","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_16K","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-11-24","metricId":"mean_score","value":81.66666666666667,"rawValue":"0.8166666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=136;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7dc6afa1be23e6e11bf3d5ee","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_32K","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-11-24","metricId":"mean_score","value":86.11111111111111,"rawValue":"0.8611111111111112","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=135;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f3f27b561d8b5179a45f0625","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_32K","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-05","metricId":"mean_score","value":93.05555555555556,"rawValue":"0.9305555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=122;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a1b1a4726c0b113c58fb3d67","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_64K","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-05","metricId":"mean_score","value":94.44444444444444,"rawValue":"0.9444444444444444","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=121;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-80d5523c6f734f505539d352","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-05","metricId":"mean_score","value":91.11111111111111,"rawValue":"0.9111111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=54;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-e454cf618347a5ea05524550","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-16","metricId":"mean_score","value":86.66666666666667,"rawValue":"0.8666666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=70;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9c91aeb4dce6340d223b41cf","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_xhigh","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-16","metricId":"mean_score","value":97.8,"rawValue":"0.978","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=113;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-daf568826a7f6765be800cc9","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-05-28","metricId":"mean_score","value":97.77777777777777,"rawValue":"0.9777777777777777","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=69;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-fb632add60450720797bd5a0","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-05-28","metricId":"mean_score","value":98.33333333333333,"rawValue":"0.9833333333333333","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=108;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-88cc73e547df77f32c0e8f89","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-05-28","metricId":"mean_score","value":84.44444444444444,"rawValue":"0.8444444444444444","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=68;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-5e5389480102410ed0dcb145","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-24","metricId":"mean_score","value":97.77777777777777,"rawValue":"0.9777777777777777","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=67;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-2b746311d780525a0b362f9d","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-24","metricId":"mean_score","value":93.33333333333333,"rawValue":"0.9333333333333333","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=65;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-eb92fa3faf5039bc0497e02d","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-24","metricId":"mean_score","value":98.88888888888889,"rawValue":"0.9888888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=79;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6e5a2bb57c78582745eb616a","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-09-29","metricId":"mean_score","value":35.55555555555556,"rawValue":"0.35555555555555557","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=153;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-cc65443da9c10315cf43e232","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_16K","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-09-29","metricId":"mean_score","value":71.11111111111111,"rawValue":"0.7111111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=146;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-98c71544f8df6299c38d0092","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-09-29","metricId":"mean_score","value":77.77777777777779,"rawValue":"0.7777777777777778","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=150;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-8cb0f9904d39ffedc08cf48b","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_59K","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-09-29","metricId":"mean_score","value":77.77777777777779,"rawValue":"0.7777777777777778","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=145;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-bf022e6d1f3873c7c4dd9bc5","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_32K","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-17","metricId":"mean_score","value":85.7936507936508,"rawValue":"0.857936507936508","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=118;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-ea52e4f600a048b96027aa6c","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-17","metricId":"mean_score","value":75.55555555555556,"rawValue":"0.7555555555555555","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=93;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d37304988e08fdf0b07b2cfe","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-17","metricId":"mean_score","value":71.11111111111111,"rawValue":"0.7111111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=66;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-73be04a8b7af18adde0020cf","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_medium","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-17","metricId":"mean_score","value":82.22222222222221,"rawValue":"0.8222222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=96;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-ca24ba78de92cdf6e4b4cdf7","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-06-30","metricId":"mean_score","value":80.0,"rawValue":"0.8","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=64;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-175bc348874a5d0ef050d8eb","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_xhigh","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-06-30","metricId":"mean_score","value":94.72222222222221,"rawValue":"0.9472222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=103;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-dc5c043f359719726646f92f","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-chat","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-12-01","metricId":"mean_score","value":48.888888888888886,"rawValue":"0.4888888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=83;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c547176c94bc5a0de9075701","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-31","metricId":"mean_score","value":94.44444444444444,"rawValue":"0.9444444444444444","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=77;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9634ca508e76139e97ccd735","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-08-13","metricId":"mean_score","value":98.61111111111111,"rawValue":"0.9861111111111112","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=3;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bac08a00cd22779cd1ef7b7c","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-24","metricId":"mean_score","value":95.55555555555556,"rawValue":"0.9555555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=53;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-888e3200392d553ed882bdf9","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-24","metricId":"mean_score","value":96.66666666666667,"rawValue":"0.9666666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=106;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-63c1c73d826938ce1d891ca2","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-24","metricId":"mean_score","value":46.666666666666664,"rawValue":"0.4666666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=52;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-1a82c1a950899d1f91643742","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"fireworks/kimi-k2p5","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-01-27","metricId":"mean_score","value":92.2,"rawValue":"0.922","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=123;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3a98447da8b1bf20905ead5c","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-06-17","metricId":"mean_score","value":84.16666666666667,"rawValue":"0.8416666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=139;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-067ec002cadb837fdba17c31","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-19","metricId":"mean_score","value":95.6,"rawValue":"0.956","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=119;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-0fe00ffe7de86505206c55b8","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-19","metricId":"mean_score","value":95.55555555555556,"rawValue":"0.9555555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=51;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6867a313c64a644b5b7e63b5","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":71.11111111111111,"rawValue":"0.7111111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=58;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-39aeba4be7bdee7371ac9c66","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":60.0,"rawValue":"0.6","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=56;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-094690c94be27826d252fc21","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_minimal","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":51.11111111111111,"rawValue":"0.5111111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=57;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-aa03db29fb19e0819bad75de","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-05-19","metricId":"mean_score","value":95.55555555555556,"rawValue":"0.9555555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=109;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-538314937c2733496441efad","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-05-19","metricId":"mean_score","value":88.88888888888889,"rawValue":"0.8888888888888888","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=59;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-32ee25b28bf7363bcac8ebde","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_minimal","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-05-19","metricId":"mean_score","value":80.0,"rawValue":"0.8","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=91;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-acad1a7307fd5da6c373e34f","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":94.16666666666667,"rawValue":"0.9416666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=78;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9eedbcd873e5772d174140e5","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":82.22222222222221,"rawValue":"0.8222222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=36;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-27619cdcdd406ce3b83ed3be","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_minimal","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-21","metricId":"mean_score","value":80.0,"rawValue":"0.8","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=35;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-2609b588181aa78b453b7a83","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-08-13","metricId":"mean_score","value":97.22222222222221,"rawValue":"0.9722222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=4;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-27c3401a344bb4ae8f3b170e","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-12-22","metricId":"mean_score","value":83.33333333333334,"rawValue":"0.8333333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=124;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8da2c7d979b433f5ea986167","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-11","metricId":"mean_score","value":80.0,"rawValue":"0.8","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=120;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2316afa4ae5f4d0bd7fe8f72","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-07","metricId":"mean_score","value":93.33333333333333,"rawValue":"0.9333333333333333","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=10;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-64529557d3137cd182518bb7","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-06-16","metricId":"mean_score","value":86.38888888888889,"rawValue":"0.8638888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=104;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-f934db767eeddfe7ae88d228","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-06-16","metricId":"mean_score","value":28.888888888888886,"rawValue":"0.28888888888888886","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=9;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-905694388eb16ed126f05d6b","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-08-14","metricId":"mean_score","value":91.11111111111111,"rawValue":"0.9111111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=2;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2fcb5614d7e99e2b82d9fb3e","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":91.38888888888889,"rawValue":"0.9138888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=144;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-de8b835dc9482adf66e5bcd8","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":87.22222222222223,"rawValue":"0.8722222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=154;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-53a8fdfff18da7b7a154d2e8","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_minimal","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":46.666666666666664,"rawValue":"0.4666666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=81;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-a43e02d5fb4e1725ac9b7c4c","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":86.66666666666667,"rawValue":"0.8666666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=143;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-b8dc75cd7b4e06fc1875dea2","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":78.33333333333333,"rawValue":"0.7833333333333333","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=155;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d01c5826cc808dbd93c84d13","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_minimal","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":55.55555555555556,"rawValue":"0.5555555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=48;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0c21237eea6749c9707ed375","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":81.11111111111111,"rawValue":"0.8111111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=142;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-436c026dea6e34226ffdba17","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":46.666666666666664,"rawValue":"0.4666666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=92;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-963eb5d76022e43542fa4bf5","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":74.16666666666667,"rawValue":"0.7416666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=156;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b679b26388401b5e5b96f014","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_minimal","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-08-07","metricId":"mean_score","value":35.55555555555556,"rawValue":"0.35555555555555557","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=94;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8013d5e8c7cc0cb2f69ff5a5","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-11-13","metricId":"mean_score","value":88.61111111111111,"rawValue":"0.8861111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=140;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-8d42547490f97eeca5403afa","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-11-13","metricId":"mean_score","value":63.888888888888886,"rawValue":"0.6388888888888888","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=133;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-d6d16b9e45558d610b250b7e","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-11-13","metricId":"mean_score","value":85.55555555555556,"rawValue":"0.8555555555555555","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=138;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-94efb4f3d917f1822cf59d88","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-11-13","metricId":"mean_score","value":37.77777777777778,"rawValue":"0.37777777777777777","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=23;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f58fb5201078b8eeff809b1c","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-12-11","metricId":"mean_score","value":96.11111111111111,"rawValue":"0.9611111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=130;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-11e404fd5ba034db69496290","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-12-11","metricId":"mean_score","value":78.88888888888889,"rawValue":"0.7888888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=131;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-e3d5978a2c59e0fabde151c9","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-12-11","metricId":"mean_score","value":93.88888888888889,"rawValue":"0.9388888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=129;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-693716815216496d64a2bd0e","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-12-11","metricId":"mean_score","value":62.22222222222222,"rawValue":"0.6222222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=95;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-7e19c18b0f768ff458e408a5","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2025-12-11","metricId":"mean_score","value":96.1,"rawValue":"0.961","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=127;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ec87d47b6033e2f40c65f27f","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":97.77777777777777,"rawValue":"0.9777777777777777","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=89;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-edf993eef5d2da52be012b83","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":84.44444444444444,"rawValue":"0.8444444444444444","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=90;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7202fe4cce7d9a5cb07c59c2","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_medium","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":95.55555555555556,"rawValue":"0.9555555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=87;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-8db681b650d7d3d914555127","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":57.77777777777777,"rawValue":"0.5777777777777777","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=88;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-2b356bbdc758fed09fce5f5c","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-05","metricId":"mean_score","value":95.3,"rawValue":"0.953","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=117;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-9907c8a71eed8446c62a7975","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":87.22222222222223,"rawValue":"0.8722222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=114;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4cd8ce94901bf53274cd94ff","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":26.666666666666668,"rawValue":"0.26666666666666666","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=17;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-0031bab9d18ff42a11def72a","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_xhigh","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":88.88888888888889,"rawValue":"0.8888888888888888","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=11;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-43accd713f83ff77a5ca3600","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":87.78,"rawValue":"0.8778","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=115;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-c625a5137a0999271e095dc9","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":68.88888888888889,"rawValue":"0.6888888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=49;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ba483ffa69cd3ee79f874dbf","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-17","metricId":"mean_score","value":46.666666666666664,"rawValue":"0.4666666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=45;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e0c720a723096512907a6f90","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-05-05","metricId":"mean_score","value":68.05555555555556,"rawValue":"0.6805555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=76;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3efb2755e63833ca444e4ad3","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-23","metricId":"mean_score","value":84.44444444444444,"rawValue":"0.8444444444444444","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=20;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-81b3ce08a7c9db0a0498ea9c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-23","metricId":"mean_score","value":57.77777777777777,"rawValue":"0.5777777777777777","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=13;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-35a9b6f94936dc7be4556d23","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":66.66666666666666,"rawValue":"0.6666666666666666","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=43;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-267e796f9cc784b8e1d9b562","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":98.33333333333333,"rawValue":"0.9833333333333333","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=100;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-a6771c14c16992760ae2efce","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":40.0,"rawValue":"0.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=27;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8be56ab7d484fcd3121bfeeb","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":95.55555555555556,"rawValue":"0.9555555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=19;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8874657f7e6840d5f9d8b0a9","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":100.0,"rawValue":"1.0","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=101;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-5ba690f6c10cbd4c90ac1005","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":68.88888888888889,"rawValue":"0.6888888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=12;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-dbce9afd188a70e401a9d340","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":88.88888888888889,"rawValue":"0.8888888888888888","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=26;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3391ac7ae7a8fb58245c939b","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":99.72222222222223,"rawValue":"0.9972222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=99;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-3b128300b701e36bf0d433f1","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-09","metricId":"mean_score","value":53.333333333333336,"rawValue":"0.5333333333333333","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=16;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-90e98b8a71a4d27b52f313b4","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-08","metricId":"mean_score","value":97.77777777777777,"rawValue":"0.9777777777777777","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=102;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cd030b489b3d9cfc98043d8a","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-08-12","metricId":"mean_score","value":97.77777777777777,"rawValue":"0.9777777777777777","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=7;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-13b562b27604e4751a0c45d8","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_xhigh","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-08-12","metricId":"mean_score","value":99.16666666666667,"rawValue":"0.9916666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=6;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ecc0809b8f799db7e9615026","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-20","metricId":"mean_score","value":96.11111111111111,"rawValue":"0.9611111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=110;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-399c54d737e634200c6c0254","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-06-12","metricId":"mean_score","value":95.55555555555556,"rawValue":"0.9555555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=25;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-694898711bdff189d3443d30","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_high","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-16","metricId":"mean_score","value":93.33333333333333,"rawValue":"0.9333333333333333","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=15;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-e7e099e68743e95ae97b5da0","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_low","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-16","metricId":"mean_score","value":68.88888888888889,"rawValue":"0.6888888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=14;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-fce64a24aaaa409eee2f865b","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-16","metricId":"mean_score","value":97.22222222222221,"rawValue":"0.9722222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=82;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8d007c49245b9a5d548578d5","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-13","metricId":"mean_score","value":88.88888888888889,"rawValue":"0.8888888888888888","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=32;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-be5bb34b23c7647666a266f0","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-13","metricId":"mean_score","value":82.22222222222221,"rawValue":"0.8222222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=33;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6dd93a1432f838d6853cc526","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-25","metricId":"mean_score","value":84.44444444444444,"rawValue":"0.8444444444444444","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=47;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-62a86ddb7a506a93a1ab8598","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-02-16","metricId":"mean_score","value":86.66666666666667,"rawValue":"0.8666666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=34;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c5e8ac747958e3b58545e323","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-22","metricId":"mean_score","value":91.11111111111111,"rawValue":"0.9111111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=30;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e45d564b054731b6d062bba9","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-22","metricId":"mean_score","value":66.66666666666666,"rawValue":"0.6666666666666666","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=31;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f43413a03173147865216cd9","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-14","metricId":"mean_score","value":86.66666666666667,"rawValue":"0.8666666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=37;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-054aa2b5315890d555728ea0","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-14","metricId":"mean_score","value":68.88888888888889,"rawValue":"0.6888888888888889","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=38;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-6ef297144f96c91964a9bbe2","canonicalModelId":"qwen/qwen3.6-flash@2026-04-27","modelName":"Qwen3.6 Flash","modelRef":"qwen3.6-flash","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-27","metricId":"mean_score","value":84.44444444444444,"rawValue":"0.8444444444444444","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=44;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cd79bb0560d02e681c591ef0","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-04-20","metricId":"mean_score","value":91.11111111111111,"rawValue":"0.9111111111111111","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=18;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a894c155cfc85f2e7516e20e","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-03-31","metricId":"mean_score","value":93.33333333333333,"rawValue":"0.9333333333333333","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=28;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c7d6dbea04100cc1e219c926","canonicalModelId":"qwen/qwen3.7-flash@2026-07-27","modelName":"Qwen3.7 Flash","modelRef":"qwen3.7-flash","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-27","metricId":"mean_score","value":86.66666666666667,"rawValue":"0.8666666666666667","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=41;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-db633fbd156f5289864c9e8e","canonicalModelId":"qwen/qwen3.7-flash@2026-07-27","modelName":"Qwen3.7 Flash","modelRef":"qwen3.7-flash_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-07-27","metricId":"mean_score","value":77.77777777777779,"rawValue":"0.7777777777777778","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=42;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-22b047319331640a35fd0cbb","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-05-19","metricId":"mean_score","value":95.55555555555556,"rawValue":"0.9555555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=21;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-54c1f6677f76521584670e9c","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-06-02","metricId":"mean_score","value":93.33333333333333,"rawValue":"0.9333333333333333","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=39;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-dc6731c3d06cccab80af901a","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus_none","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-06-02","metricId":"mean_score","value":80.0,"rawValue":"0.8","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=40;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-05408cbe684e639d7f2a3057","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max_xhigh","benchmarkId":"epoch-otis_mock_aime_2024_2025","benchmarkVersion":"2026-08-02","metricId":"mean_score","value":99.44444444444444,"rawValue":"0.9944444444444445","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=75;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4ec6b0216068cf21987dab45","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-02-12","metricId":"Accuracy","value":4.0,"rawValue":"0.04","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=55;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8674c2b6e0b537761bde5d9d","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMax-M2.7","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-03-18","metricId":"Accuracy","value":3.0,"rawValue":"0.03","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=56;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-512617e1a7c5197adae457c8","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-06-01","metricId":"Accuracy","value":18.0,"rawValue":"0.18","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=34;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e97fcb3138af24d398cb8ad1","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-06-09","metricId":"Accuracy","value":95.0,"rawValue":"0.9500000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=3;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e10add905c59b6ac89008d7a","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy","value":36.0,"rawValue":"0.36","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=22;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-60e462b5b1abe0cb153e7f44","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy","value":50.0,"rawValue":"0.5","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=17;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b7aa0140236231a726842dc4","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-04-16","metricId":"Accuracy","value":54.0,"rawValue":"0.54","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=14;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4cb9977a0ad1bdb61d274d82","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-05-28","metricId":"Accuracy","value":69.0,"rawValue":"0.6900000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=8;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-08cb5e8af20a4bf7ab7cb49f","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-07-24","metricId":"Accuracy","value":99.0,"rawValue":"0.99","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=2;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-58cfc93cf5ec29cb12feede4","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy","value":19.0,"rawValue":"0.19","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=33;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0b251da9762836f5e1d8cafb","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-02-17","metricId":"Accuracy","value":45.0,"rawValue":"0.45","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=19;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-df405c3a103ab89c8dac4d18","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-06-30","metricId":"Accuracy","value":77.0,"rawValue":"0.77","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=6;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f3bac702baf74d4e4bfff126","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_unknown","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-07-31","metricId":"Accuracy","value":56.00000000000001,"rawValue":"0.56","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=13;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-99833c2a62a070b6c1341942","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_unknown","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-08-13","metricId":"Accuracy","value":50.0,"rawValue":"0.5","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=18;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-24a7011ea18374c5adabdcab","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-04-24","metricId":"Accuracy","value":16.0,"rawValue":"0.16","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=39;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c501734a6ba9999134d2ea36","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"fireworks/deepseek-v3p2","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2025-12-01","metricId":"Accuracy","value":8.0,"rawValue":"0.08","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=50;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-37997489cdaf35c0022efbcd","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-02-19","metricId":"Accuracy","value":26.0,"rawValue":"0.26","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=28;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4ce248f17242102156eca0d0","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-07-21","metricId":"Accuracy","value":13.0,"rawValue":"0.13","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=44;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6bb3bda4c2ea6b1a97533d9e","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-05-19","metricId":"Accuracy","value":31.0,"rawValue":"0.31","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=26;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-75c03890f4b8b04aaf60d4b7","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_unknown","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-07-21","metricId":"Accuracy","value":36.0,"rawValue":"0.36","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=23;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-56a61a310be1afa658ffcdb4","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_unknown","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-08-13","metricId":"Accuracy","value":57.99999999999999,"rawValue":"0.58","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=10;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-768fa09a655339e98628655d","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2025-12-22","metricId":"Accuracy","value":6.0,"rawValue":"0.06","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=51;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2ce28d963ab855a69e3bf3e9","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-04-07","metricId":"Accuracy","value":22.222,"rawValue":"0.22222000000000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=29;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3ad13f3dad891ca9df217233","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-06-16","metricId":"Accuracy","value":35.0,"rawValue":"0.35000000000000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=24;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ef34e011f55f9aefffa1dd17","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":18.0,"rawValue":"0.18","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=35;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-32860098046de98e40ca30c0","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":9.0,"rawValue":"0.09","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=47;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-aa5549e3efd33b6ce01746fd","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":12.0,"rawValue":"0.12","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=45;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3b4f1674996b22db57ba5614","canonicalModelId":"openai/gpt-5.1-codex-max@2025-11-19","modelName":"GPT-5.1 Codex Max","modelRef":"gpt-5.1-codex-max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2025-11-19","metricId":"Accuracy","value":9.0,"rawValue":"0.09","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=48;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-387a59129d3f18a25a5a3317","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy","value":15.0,"rawValue":"0.15","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=42;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4688e8ccf49f5c78711797fc","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-03-05","metricId":"Accuracy","value":56.00000000000001,"rawValue":"0.56","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=12;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fcc36daa52c66917db3c043d","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_xhigh","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-03-17","metricId":"Accuracy","value":21.0,"rawValue":"0.21","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=31;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d6cc9bacdc5355e5b27cb5ea","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-03-17","metricId":"Accuracy","value":5.0,"rawValue":"0.05","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=53;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-eac83511378a5e9c44f27eec","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy","value":50.0,"rawValue":"0.5","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=16;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-dbca2ae7fcbb852b9a53783e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":60.0,"rawValue":"0.6","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=9;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c9bcf6058572ed75ecb1ca28","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":83.0,"rawValue":"0.8300000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=5;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4d813e5664865f067d80d89d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":74.0,"rawValue":"0.74","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=7;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-377814eaf02a6bd75d1867c2","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-07-08","metricId":"Accuracy","value":31.0,"rawValue":"0.31","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=25;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4d957ce1e33792febd47350f","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_unknown","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-08-12","metricId":"Accuracy","value":51.0,"rawValue":"0.51","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=15;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4c9c2b81af41927c9e190e49","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-04-20","metricId":"Accuracy","value":16.0,"rawValue":"0.16","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=37;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-987ea620e694f5a2f7b1c65c","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_unknown","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-07-16","metricId":"Accuracy","value":87.0,"rawValue":"0.87","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=4;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-30df61bba3d14af4e1a2d3c9","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"epoch-proofbench_external","metricId":"Accuracy","value":16.0,"rawValue":"0.16","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=38;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-22f7fdba41669845335c06d6","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"epoch-proofbench_external","metricId":"Accuracy","value":22.0,"rawValue":"0.22","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=30;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2f01fb23e1258d78e4a27a0e","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-05-19","metricId":"Accuracy","value":26.0,"rawValue":"0.26","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=27;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-731ed380ec13a281507dcf9d","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-08-14","metricId":"Accuracy","value":16.0,"rawValue":"0.16","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=40;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7be97d382a6edf94033b926d","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max_unknown","benchmarkId":"epoch-proofbench_external","benchmarkVersion":"2026-08-02","metricId":"Accuracy","value":57.99999999999999,"rawValue":"0.58","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=11;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5df68ad1cbe2919308bd0d59","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"epoch-rli_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.161,"rawValue":"0.161","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"rli_external.csv:row=2;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-b43fda2c6335e1031eb14380","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_unknown","benchmarkId":"epoch-rli_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.158,"rawValue":"0.158","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"rli_external.csv:row=3;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d80275c82402326edab6c0ec","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-rli_external","benchmarkVersion":"2025-11-24","metricId":"Score","value":0.0375,"rawValue":"0.0375","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"rli_external.csv:row=8;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ca7ead15c905342e259b552a","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-rli_external","benchmarkVersion":"2026-02-05","metricId":"Score","value":0.0417,"rawValue":"0.0417","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"rli_external.csv:row=7;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f38db8e873eba2389f842deb","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_unknown","benchmarkId":"epoch-rli_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.0833,"rawValue":"0.0833","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"rli_external.csv:row=4;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-23488cd0590347ccbfe3c0bd","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-rli_external","benchmarkVersion":"2025-09-29","metricId":"Score","value":0.020800000000000003,"rawValue":"0.020800000000000003","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"rli_external.csv:row=11;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9764ef857872ee056150b114","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_unknown","benchmarkId":"epoch-rli_external","benchmarkVersion":"2026-08-13","metricId":"Score","value":0.05,"rawValue":"0.05","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"rli_external.csv:row=6;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3db575785bc64c819218207d","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_unknown","benchmarkId":"epoch-rli_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.0167,"rawValue":"0.0167","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"rli_external.csv:row=12;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-db5fc514c2a98bde04323a97","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-rli_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.025019999999999997,"rawValue":"0.025019999999999997","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"rli_external.csv:row=9;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0a15bdd2489f6b5ed176cf7c","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-rli_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":0.020800000000000003,"rawValue":"0.020800000000000003","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"rli_external.csv:row=10;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-356a912c3fc8d24b9b11233f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-rli_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.0625,"rawValue":"0.0625","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"rli_external.csv:row=5;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-616b4c2fbd71bd23cd5a388d","canonicalModelId":"deepseek/v3.1-terminus@2025-09-22","modelName":"DeepSeek V3.1 Terminus","modelRef":"DeepSeek-V3.1-Terminus","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2025-09-22","metricId":"Score","value":0.40625,"rawValue":"0.40625","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=123;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d3a3ea78f7d19c225de00738","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp_thinking","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2025-09-29","metricId":"Score","value":0.388888888888889,"rawValue":"0.388888888888889","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=126;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-763461e4b0d5a502b434ec76","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMax-M2.7","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-03-18","metricId":"Score","value":0.469907407407407,"rawValue":"0.469907407407407","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=22;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0a9970aca32608b44d90efa7","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-06-01","metricId":"Score","value":0.45370370370370405,"rawValue":"0.45370370370370405","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=25;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4537ef42d26b9327ff19f705","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":0.601851851851852,"rawValue":"0.601851851851852","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=2;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0177a0a1f4f50b5178167151","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2025-10-15","metricId":"Score","value":0.43287037037037,"rawValue":"0.43287037037037","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=29;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c41567a6e2d968a9c9b18efd","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":0.545138888888889,"rawValue":"0.545138888888889","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=7;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f9a19f173cc1ff7b28cd1221","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":0.534722222222222,"rawValue":"0.534722222222222","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=9;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c23e7c4dd574d3511a7f72ed","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_high","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.542824074074074,"rawValue":"0.542824074074074","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=74;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-a9563121e91bc85ca3b1af3e","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_low","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.48032407407407407,"rawValue":"0.48032407407407407","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=76;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-c87159f3ae2a28622ca3955d","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.556712962962963,"rawValue":"0.556712962962963","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=72;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-2e14ccc76ae5c30e4a9e6842","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_medium","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.506944444444444,"rawValue":"0.506944444444444","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=75;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-db6c17027247f8544db3a3b8","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_xhigh","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":0.549768518518518,"rawValue":"0.549768518518518","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=73;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bfb95126f3949b769ae2feca","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2025-09-29","metricId":"Score","value":0.446759259259259,"rawValue":"0.446759259259259","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=120;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1f31ac5a6ff5fc7b975a1155","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-02-17","metricId":"Score","value":0.467592592592593,"rawValue":"0.467592592592593","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=24;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0cbff5a92553636711f475c1","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_high","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-06-30","metricId":"Score","value":0.486111111111111,"rawValue":"0.486111111111111","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=114;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-23b28249669c066f0236e2a0","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-06-30","metricId":"Score","value":0.53587962962963,"rawValue":"0.53587962962963","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=39;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0cd979c5f28c081adc2bd851","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-31","metricId":"Score","value":0.49884259259259306,"rawValue":"0.49884259259259306","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=111;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-eb6b6a8297942b2c83e39699","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash_high","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-24","metricId":"Score","value":0.420138888888889,"rawValue":"0.420138888888889","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=122;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-79fb22233c738ca88addd7c5","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-24","metricId":"Score","value":0.44907407407407407,"rawValue":"0.44907407407407407","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=119;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-4157a0ac33dab79fa29f536f","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-24","metricId":"Score","value":0.44907407407407407,"rawValue":"0.44907407407407407","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=26;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-c444af4809f98e72abf65342","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_high","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-24","metricId":"Score","value":0.46412037037037,"rawValue":"0.46412037037037","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=116;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c10b0a6344757c6feaddd057","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-24","metricId":"Score","value":0.5,"rawValue":"0.5","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=16;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-6369c2f040f933ba9a27ef3e","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2025-06-17","metricId":"Score","value":0.428240740740741,"rawValue":"0.428240740740741","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=41;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-987b4314cf2762f76f8d5bc3","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-02-19","metricId":"Score","value":0.58912037037037,"rawValue":"0.58912037037037","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=3;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3d7f629e3e10caf9bb8ff6de","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.40856481481481505,"rawValue":"0.40856481481481505","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=87;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-92d35b1fcb3657d91bbfae4d","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-05-19","metricId":"Score","value":0.53125,"rawValue":"0.53125","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=11;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c6dca6fe0cdf6bdb68e3e6bf","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-21","metricId":"Score","value":0.52662037037037,"rawValue":"0.52662037037037","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=79;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c3ca8883d3f05d8d2a273b53","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2025-12-22","metricId":"Score","value":0.451388888888889,"rawValue":"0.451388888888889","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=118;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cc00ad4fd9743b618f42d6e5","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-07","metricId":"Score","value":0.4375,"rawValue":"0.4375","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=27;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3bdfd7aab1de3dd1a347a67b","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-06-16","metricId":"Score","value":0.50462962962963,"rawValue":"0.50462962962963","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=14;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-f930618e8eb878544b23b321","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_none","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-06-16","metricId":"Score","value":0.361111111111111,"rawValue":"0.361111111111111","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=129;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7c6cf1ba6563087310d050aa","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_unknown","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.429398148148148,"rawValue":"0.429398148148148","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=70;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-318dbd41f5a8ce782f5424ff","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":0.392361111111111,"rawValue":"0.392361111111111","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=71;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4cb2fe74a436860d27bba535","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2025-11-13","metricId":"Score","value":0.43287037037037,"rawValue":"0.43287037037037","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=53;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7d90fa7d53f0777e021907a3","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-03-05","metricId":"Score","value":0.565972222222222,"rawValue":"0.565972222222222","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=4;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-027151496ae6d5438354666b","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_xhigh","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.49884259259259306,"rawValue":"0.49884259259259306","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=17;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-45116879f24886c1c090a746","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_xhigh","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-03-17","metricId":"Score","value":0.46875,"rawValue":"0.46875","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=23;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dd7a22656f19e7866ecd8a3f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-05-05","metricId":"Score","value":0.486111111111111,"rawValue":"0.486111111111111","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=40;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-16ebec81af888f540d5a57a0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.559027777777778,"rawValue":"0.559027777777778","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=6;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-99e05655bb6b4f4d954c8cc9","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_low","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.516203703703704,"rawValue":"0.516203703703704","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=12;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-ffc26fca3cf6257220979588","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_medium","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.534722222222222,"rawValue":"0.534722222222222","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=8;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ca390ed0960b288fcf8d8375","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_none","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.47337962962963,"rawValue":"0.47337962962963","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=20;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-c9c2448dc803829f02fc5eaf","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":0.561342592592593,"rawValue":"0.561342592592593","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=5;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-f9a97a1965455820c20ebb42","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_high","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.506944444444444,"rawValue":"0.506944444444444","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=63;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d7dc8307fa0cfbeea3a79c0c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_low","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.456018518518519,"rawValue":"0.456018518518519","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=69;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b244001850b59ec525c82432","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.525462962962963,"rawValue":"0.525462962962963","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=61;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-0efb504b19f64914fe365836","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_medium","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.45833333333333304,"rawValue":"0.45833333333333304","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=68;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-8e57ef996a14d2b9d6a54a9c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_none","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.399305555555556,"rawValue":"0.399305555555556","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=124;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-af32d951b3c4ee9682eb799c","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_xhigh","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.5,"rawValue":"0.5","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=65;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-93c3ff1ebf2781046d1a18b8","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_high","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.569444444444444,"rawValue":"0.569444444444444","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=54;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-e862afc99dac711001d9d2ad","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_low","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.554398148148148,"rawValue":"0.554398148148148","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=58;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-40a98bae325d28a744b644ea","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.561342592592593,"rawValue":"0.561342592592593","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=56;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-75d3004fe62ca78e62d5e76b","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_medium","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.564814814814815,"rawValue":"0.564814814814815","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=55;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-6519d86647c43ebbe2a4e9b7","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_none","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.47106481481481505,"rawValue":"0.47106481481481505","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=115;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-f839632b8907bcf7396f875e","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.560185185185185,"rawValue":"0.560185185185185","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=57;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0910929be26b898d650631d2","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_high","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.501157407407407,"rawValue":"0.501157407407407","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=64;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":6},{"id":"pub-332ce289161761d15eca2353","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_low","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.491898148148148,"rawValue":"0.491898148148148","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=67;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-41499dbb654200d8bbe407af","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.539351851851852,"rawValue":"0.539351851851852","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=60;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-7f7aaf5c3e80cf9ca025f5a2","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_medium","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.496527777777778,"rawValue":"0.496527777777778","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=66;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-f9b607fd51d46581bdeb8813","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_none","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.4456018518518521,"rawValue":"0.4456018518518521","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=121;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3b9e9a78dc083dc22f91f4a9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":0.516203703703704,"rawValue":"0.516203703703704","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=62;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-8a2a0ee33e047dd257bd773d","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-08","metricId":"Score","value":0.540509259259259,"rawValue":"0.540509259259259","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=59;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-21b14429ef6c7ee3fa5893e3","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-01-27","metricId":"Score","value":0.48958333333333304,"rawValue":"0.48958333333333304","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=112;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4f9ce973267313e6c5887a8d","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-20","metricId":"Score","value":0.534722222222222,"rawValue":"0.534722222222222","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=10;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-10d700b25a4a586197be0784","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-06-12","metricId":"Score","value":0.474537037037037,"rawValue":"0.474537037037037","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=19;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-bf15508df436876d594b38f2","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_low","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-16","metricId":"Score","value":0.511574074074074,"rawValue":"0.511574074074074","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=110;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-33008d65d61ef036e16d4cca","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-16","metricId":"Score","value":0.586805555555556,"rawValue":"0.586805555555556","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=109;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-5ef6987a16ff9b3364c51acb","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_unknown","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-07-16","metricId":"Score","value":0.586805555555556,"rawValue":"0.586805555555556","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=77;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-cebe323335e47504a6c3c551","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash","benchmarkId":"epoch-scicode_external","metricId":"Score","value":0.259259259259259,"rawValue":"0.259259259259259","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=49;column=Score","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c52760e0867b6b589b44f841","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"epoch-scicode_external","metricId":"Score","value":0.430555555555556,"rawValue":"0.430555555555556","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=82;column=Score","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2d5b9394aeacad3574c8aa0d","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"epoch-scicode_external","metricId":"Score","value":0.502314814814815,"rawValue":"0.502314814814815","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=15;column=Score","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9d3f3cde6caa5e374830283b","canonicalModelId":"qwen/qwen3-coder-next@2026-02-03","modelName":"Qwen3-Coder-Next","modelRef":"qwen3-coder-next","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-02-02","metricId":"Score","value":0.322916666666667,"rawValue":"0.322916666666667","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=99;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-012188073aac34d35b101063","canonicalModelId":"qwen/qwen3.5-9b@current","modelName":"Qwen3.5 9B","modelRef":"qwen3.5-9B","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-02-24","metricId":"Score","value":0.275462962962963,"rawValue":"0.275462962962963","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=47;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f9a6170106f104b92772ef33","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-04-14","metricId":"Score","value":0.357638888888889,"rawValue":"0.357638888888889","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=35;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-06fcd0284e4582bbfe808a25","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-03-31","metricId":"Score","value":0.407407407407407,"rawValue":"0.407407407407407","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=31;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-af0bb3240c4f1f82a03e71dc","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-05-19","metricId":"Score","value":0.488425925925926,"rawValue":"0.488425925925926","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=18;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-da5125d3d709c6828b9f2c35","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"epoch-scicode_external","benchmarkVersion":"2026-06-02","metricId":"Score","value":0.454861111111111,"rawValue":"0.454861111111111","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"scicode_external.csv:row=81;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a9cf66f2bccbd7128419b84e","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-05-28","metricId":"Score (AVG@5)","value":0.408,"rawValue":"0.408","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=73;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e4106f72db561ed2d141772b","canonicalModelId":"deepseek/v3.1@2025-08-21","modelName":"DeepSeek V3.1","modelRef":"DeepSeek-V3.1","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-08-21","metricId":"Score (AVG@5)","value":0.4,"rawValue":"0.4","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=75;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7673f3ac4617acf783dc3b62","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-06-01","metricId":"Score (AVG@5)","value":0.45799999999999996,"rawValue":"0.45799999999999996","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=66;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-db742de79d6482ba46b437e3","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-06-09","metricId":"Score (AVG@5)","value":0.8190000000000001,"rawValue":"0.8190000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=2;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-802a920e95ea36d0c59e925a","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_12K","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-05-22","metricId":"Score (AVG@5)","value":0.588,"rawValue":"0.588","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=35;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5b9c6b5369d3a7baad94e59b","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-05-22","metricId":"Score (AVG@5)","value":0.588,"rawValue":"0.588","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=36;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-6b7af66850e6835a08bb8ec2","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-11-24","metricId":"Score (AVG@5)","value":0.62,"rawValue":"0.62","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=23;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3a21bd23887ea5f4a95f9b64","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-11-24","metricId":"Score (AVG@5)","value":0.62,"rawValue":"0.62","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=24;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e58df336e36cbb20897a0ee4","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-02-05","metricId":"Score (AVG@5)","value":0.6759999999999999,"rawValue":"0.6759999999999999","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=17;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f0ab10d4ca225b56457ce118","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-04-16","metricId":"Score (AVG@5)","value":0.617,"rawValue":"0.617","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=25;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5bc2feeaf56135113ecc56f2","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-05-28","metricId":"Score (AVG@5)","value":0.648,"rawValue":"0.648","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=18;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7b57dfdcfbcea1eba9b77282","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-24","metricId":"Score (AVG@5)","value":0.8059999999999999,"rawValue":"0.8059999999999999","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=3;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0501cf34b8116ce826e3f4bb","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_12K","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-09-29","metricId":"Score (AVG@5)","value":0.543,"rawValue":"0.543","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=44;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ba7ee41c4e58f6f0329b8727","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-09-29","metricId":"Score (AVG@5)","value":0.543,"rawValue":"0.543","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=45;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d1004bd1a3758e2d931aa274","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-06-30","metricId":"Score (AVG@5)","value":0.606,"rawValue":"0.606","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=31;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d3b6dfd0e85414e19071a065","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-31","metricId":"Score (AVG@5)","value":0.611,"rawValue":"0.611","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=29;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b1b02207cd2d54292343d9c8","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-04-24","metricId":"Score (AVG@5)","value":0.46299999999999997,"rawValue":"0.46299999999999997","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=63;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-7febb86b620e8676f56c7cb5","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-06-17","metricId":"Score (AVG@5)","value":0.412,"rawValue":"0.412","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=72;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-26c2f0f653c7ca82504e4fcb","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-02-19","metricId":"Score (AVG@5)","value":0.7959999999999999,"rawValue":"0.7959999999999999","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=4;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ffe698f82f9f55cac92c8a3a","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-05-19","metricId":"Score (AVG@5)","value":0.767,"rawValue":"0.767","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=7;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4158a519a8aa708bcddab697","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-12-22","metricId":"Score (AVG@5)","value":0.477,"rawValue":"0.477","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=57;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-639b4bd168029c488219a104","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-02-11","metricId":"Score (AVG@5)","value":0.532,"rawValue":"0.532","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=47;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a7d91f5f186d5f99bb84eec2","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-04-07","metricId":"Score (AVG@5)","value":0.551,"rawValue":"0.551","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=43;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a5fc187883b333a4413b9080","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-06-16","metricId":"Score (AVG@5)","value":0.588,"rawValue":"0.588","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=37;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6419ac69436044e056b5fc91","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-08-07","metricId":"Score (AVG@5)","value":0.5670000000000001,"rawValue":"0.5670000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=40;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ed76eb76f23627038512304f","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-11-13","metricId":"Score (AVG@5)","value":0.532,"rawValue":"0.532","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=46;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f2e03803274e33f813396aa1","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-12-11","metricId":"Score (AVG@5)","value":0.458,"rawValue":"0.458","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=64;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-68e156028ba3347791c4e72c","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-12-11","metricId":"Score (AVG@5)","value":0.45799999999999996,"rawValue":"0.45799999999999996","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=65;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-a930b31d0d13c4c1ff8d0b68","canonicalModelId":"openai/gpt-5.2-pro@2025-12-11","modelName":"GPT-5.2 Pro","modelRef":"gpt-5.2-pro-2025-12-11","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-12-11","metricId":"Score (AVG@5)","value":0.574,"rawValue":"0.574","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=39;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0e1f71882a15a46cc5002c8d","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-03-05","metricId":"Score (AVG@5)","value":0.741,"rawValue":"0.741","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=11;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2966eaa566be6508c8ee8dc1","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_high","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-04-23","metricId":"Score (AVG@5)","value":0.7690000000000001,"rawValue":"0.7690000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=6;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d9266c96de3686f093f64966","canonicalModelId":"openai/gpt-5.5-pro@current","modelName":"GPT-5.5 Pro","modelRef":"gpt-5.5-pro_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-04-23","metricId":"Score (AVG@5)","value":0.7690000000000001,"rawValue":"0.7690000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=5;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c21c637ec3b23541e656ed36","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-04-23","metricId":"Score (AVG@5)","value":0.6900000000000001,"rawValue":"0.6900000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=16;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e0c6914561641a12491e5627","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-09","metricId":"Score (AVG@5)","value":0.468,"rawValue":"0.468","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=58;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0d595be343ced48d4a7bcc13","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_xhigh","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-09","metricId":"Score (AVG@5)","value":0.46799999999999997,"rawValue":"0.46799999999999997","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=60;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-973939c83c348398a527b8f4","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_prounknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-09","metricId":"Score (AVG@5)","value":0.717,"rawValue":"0.717","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=13;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-72072a9bba5f2fbcfb54ca82","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_proxhigh","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-09","metricId":"Score (AVG@5)","value":0.7170000000000001,"rawValue":"0.7170000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=12;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-d36330ef54066f3f47f0ecd4","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-09","metricId":"Score (AVG@5)","value":0.648,"rawValue":"0.648","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=19;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3f946af8ea3314044e978cc1","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-09","metricId":"Score (AVG@5)","value":0.648,"rawValue":"0.648","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=20;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-11b6bac5ee7a06778589b58d","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-09","metricId":"Score (AVG@5)","value":0.489,"rawValue":"0.489","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=54;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-6f0a5c7c9f45ee58e6060e45","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-09","metricId":"Score (AVG@5)","value":0.489,"rawValue":"0.489","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=55;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e3f0950c7b2bfa6a3dd684f0","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-08","metricId":"Score (AVG@5)","value":0.7000000000000001,"rawValue":"0.7000000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=15;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-86e6ad2ac3cba46b50a78505","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_unknown","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-08-12","metricId":"Score (AVG@5)","value":0.7590000000000001,"rawValue":"0.7590000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=9;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b6d7cf94e0dfc0f24869a91c","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-01-27","metricId":"Score (AVG@5)","value":0.46799999999999997,"rawValue":"0.46799999999999997","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=59;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9e93b1a8b2792299d95dbed3","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-06-12","metricId":"Score (AVG@5)","value":0.579,"rawValue":"0.579","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=38;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9543beb289a09ec4bd3b0b34","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-07-16","metricId":"Score (AVG@5)","value":0.6070000000000001,"rawValue":"0.6070000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=30;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-cdf6df1cdb018dc49a431c79","canonicalModelId":"qwen/qwen3.6-flash@2026-04-27","modelName":"Qwen3.6 Flash","modelRef":"qwen3.6-flash","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-04-27","metricId":"Score (AVG@5)","value":0.352,"rawValue":"0.352","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=79;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f27ec28fa512dac37c3dcec5","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-04-20","metricId":"Score (AVG@5)","value":0.63,"rawValue":"0.63","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=21;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1de0fb3e23880be31acfa7b0","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2026-05-19","metricId":"Score (AVG@5)","value":0.7040000000000001,"rawValue":"0.7040000000000001","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=14;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-baa1225e286dd2672d7ea726","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/glm-4.7","benchmarkId":"epoch-simplebench_external","benchmarkVersion":"2025-12-22","metricId":"Score (AVG@5)","value":0.47700000000000004,"rawValue":"0.47700000000000004","unit":"fraction","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=56;column=Score (AVG@5)","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-519190f15c2a5d22ac52b26c","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-spatialviz_bench_external","benchmarkVersion":"2025-06-17","metricId":"Overall score","value":36.86,"rawValue":"0.3686","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"spatialviz_bench_external.csv:row=7;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2383773e29fd156553f4c329","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-spatialviz_bench_external","benchmarkVersion":"2025-06-17","metricId":"Overall score","value":44.66,"rawValue":"0.4466","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"spatialviz_bench_external.csv:row=8;column=Overall score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-749a86d3ff20907fd4899477","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-06-01","metricId":"Mean score","value":54.99999999999999,"rawValue":"0.5499999999999999","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=15;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8750e373dfb50d073af5b216","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-06-09","metricId":"Mean score","value":95.0,"rawValue":"0.9500000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=2;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3db1d98e49fb0d5077f2039c","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_high","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-05-28","metricId":"Mean score","value":87.5,"rawValue":"0.875","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=6;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-bfe1bc146ccc8875d84ca503","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_none","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-05-28","metricId":"Mean score","value":68.125,"rawValue":"0.68125","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=12;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3f182f965584a20448a8907d","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_medium","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-06-30","metricId":"Mean score","value":60.0,"rawValue":"0.6","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=21;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-572b0de38e289df05551f606","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_high","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-04-24","metricId":"Mean score","value":40.0,"rawValue":"0.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=13;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3d50444978576bdb89954dbd","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_medium","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-05-19","metricId":"Mean score","value":58.12500000000001,"rawValue":"0.58125","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=20;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ca32a9b777d1afbad5404d39","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_high","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-06-16","metricId":"Mean score","value":55.625,"rawValue":"0.55625","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=11;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ac2a3a419961d253b40aa14f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-04-23","metricId":"Mean score","value":88.125,"rawValue":"0.88125","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=4;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-32910fdf1908134b5ac606e6","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_medium","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-04-23","metricId":"Mean score","value":81.25,"rawValue":"0.8125","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=8;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-f8df3591d07cc9ef171edd2d","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_medium","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-07-09","metricId":"Mean score","value":61.875,"rawValue":"0.61875","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=14;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-dd540ca3130e49a8ff3e6c7d","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-07-09","metricId":"Mean score","value":93.125,"rawValue":"0.93125","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=3;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-82809c08bc79f4966b0eafc9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-07-09","metricId":"Mean score","value":83.75,"rawValue":"0.8375","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=7;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9f7dbb65315c88bb5b934a99","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-07-08","metricId":"Mean score","value":74.375,"rawValue":"0.74375","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=9;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f6701530ff19e079689ba8b9","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-06-12","metricId":"Mean score","value":48.75,"rawValue":"0.4875","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=10;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-225e9cdc9d12423c20dd3e9f","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-07-16","metricId":"Mean score","value":93.0,"rawValue":"0.93","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=5;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-00898e199dd776653ca5eddf","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_unknown","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-07-16","metricId":"Mean score","value":95.0,"rawValue":"0.9500000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=19;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e3b2ad2ecce88cf7ed865fba","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-04-14","metricId":"Mean score","value":44.375,"rawValue":"0.44375000000000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=16;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-69f86c60df127c84f1f77860","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b_xhigh","benchmarkId":"epoch-surface_evolver_bench_external","benchmarkVersion":"2026-08-14","metricId":"Mean score","value":45.0,"rawValue":"0.45","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"surface_evolver_bench_external.csv:row=26;column=Mean score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4d5830019e6685cb2f714d42","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"MiniMax-M2.1","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-23","metricId":"Accuracy mean","value":29.2,"rawValue":"0.292","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=140;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-41d4f76695e499186a087c63","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"MiniMax-M2.1","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-23","metricId":"Accuracy mean","value":36.6,"rawValue":"0.36600000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=115;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Crux","subject_type":"system"},"harnessId":"Crux","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c2559f3d0d1fb4b8ada4c66f","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-12","metricId":"Accuracy mean","value":42.7,"rawValue":"0.42700000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=97;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"cchuter","subject_type":"system"},"harnessId":"cchuter","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6f62f5dffd157418e706fff7","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-12","metricId":"Accuracy mean","value":42.2,"rawValue":"0.42200000000000004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=103;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-440acb875af18653e0dc1fc2","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMax-M2.7","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-03-18","metricId":"Accuracy mean","value":45.1,"rawValue":"0.451","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=82;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"IndusAGI Coding Agent","subject_type":"system"},"harnessId":"IndusAGI Coding Agent","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-25b5cb9e37f2a50c7a413712","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMax-M2.7","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-03-18","metricId":"Accuracy mean","value":42.9,"rawValue":"0.429","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=93;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Harness Agent","subject_type":"system"},"harnessId":"Harness Agent","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-c696103f75cd0ddb4cc74433","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-07-31","metricId":"Accuracy mean","value":27.200000000000003,"rawValue":"0.272","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=150;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Dakou Agent","subject_type":"system"},"harnessId":"Dakou Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5abb260787ee9e9509203e35","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-07-31","metricId":"Accuracy mean","value":25.4,"rawValue":"0.254","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=158;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-54c3cebb93d4582b9180dda4","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-07-31","metricId":"Accuracy mean","value":23.9,"rawValue":"0.239","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=165;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-1f936642056d2fd5900a4bbf","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-07-31","metricId":"Accuracy mean","value":25.4,"rawValue":"0.254","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=160;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-09fa63b333558fac4d745177","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-07-31","metricId":"Accuracy mean","value":23.9,"rawValue":"0.239","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=166;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-c2bd7dda0cdf20d175b51ea5","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy mean","value":13.900000000000002,"rawValue":"0.139","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=191;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-bb47524979b739602c2c8ec5","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy mean","value":28.299999999999997,"rawValue":"0.283","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=143;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-ab2f8fc672c975a3b584cc07","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy mean","value":29.799999999999997,"rawValue":"0.298","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=137;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-4c0aca0e57e0ace47ea7baff","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy mean","value":27.500000000000004,"rawValue":"0.275","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=146;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-a708fad4f4ebeb48ad582b43","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy mean","value":13.900000000000002,"rawValue":"0.139","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=192;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-8caf83156ff5088560f1be06","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy mean","value":27.500000000000004,"rawValue":"0.275","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=147;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-3fba274766f2654402fd81b8","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy mean","value":35.5,"rawValue":"0.355","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=118;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Goose","subject_type":"system"},"harnessId":"Goose","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-2ca5300243b2cba45a8c5823","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy mean","value":29.799999999999997,"rawValue":"0.298","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=138;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-0de02ec4bfd66e2290021df6","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy mean","value":28.300000000000004,"rawValue":"0.28300000000000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=142;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-9d6f2c3278688c320abca4be","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy mean","value":63.1,"rawValue":"0.631","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=39;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Droid","subject_type":"system"},"harnessId":"Droid","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-083007d9b54c2450c0b5f96a","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_128K","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy mean","value":59.099999999999994,"rawValue":"0.591","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=54;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Letta Code","subject_type":"system"},"harnessId":"Letta Code","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-e0eafd8e9dfb241210c19247","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy mean","value":51.9,"rawValue":"0.519","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=71;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d61e6e0a5a23d81db1b50976","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy mean","value":58.4,"rawValue":"0.584","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=55;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mux","subject_type":"system"},"harnessId":"Mux","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b5bf0c0d509909c871b0eea0","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy mean","value":51.7,"rawValue":"0.517","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=73;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenCode","subject_type":"system"},"harnessId":"OpenCode","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-aa4f79aea5df09e7ec745c07","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy mean","value":63.1,"rawValue":"0.631","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=40;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Droid","subject_type":"system"},"harnessId":"Droid","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-6515bf6e896e6135ddf84291","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy mean","value":52.1,"rawValue":"0.521","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=70;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-647043b88d2d156e7784cbdb","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy mean","value":59.10000000000001,"rawValue":"0.5910000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=53;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Letta Code","subject_type":"system"},"harnessId":"Letta Code","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-3d6e23d4e87f7d3283bf559b","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy mean","value":57.8,"rawValue":"0.578","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=59;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-3617bcb040a06fbd52c0d5dd","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy mean","value":54.300000000000004,"rawValue":"0.543","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=64;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Goose","subject_type":"system"},"harnessId":"Goose","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-911c91830d9e4f1ba66b77fe","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":69.89999999999999,"rawValue":"0.699","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=27;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Droid","subject_type":"system"},"harnessId":"Droid","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-fee8dcb2af1ddde153600f9a","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":62.9,"rawValue":"0.629","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=43;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-06","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-face9a8bfd517360dad1f49f","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":66.5,"rawValue":"0.665","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=31;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mux","subject_type":"system"},"harnessId":"Mux","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f72bc5b6b65b276ce4903692","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":71.9,"rawValue":"0.7190000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=23;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"TongAgents","subject_type":"system"},"harnessId":"TongAgents","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-cda29aa343922d9b9940543d","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":69.9,"rawValue":"0.6990000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=26;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Droid","subject_type":"system"},"harnessId":"Droid","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-beb3865bc0301fe7d434ca38","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":74.7,"rawValue":"0.747","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=20;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus-KIRA","subject_type":"system"},"harnessId":"Terminus-KIRA","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-b81cdedfa9d5ac7e905d1157","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":57.99999999999999,"rawValue":"0.58","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=56;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-9163f767defd7063d6ebc79e","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":72.1,"rawValue":"0.721","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=22;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"MAYA-V2","subject_type":"system"},"harnessId":"MAYA-V2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-874cc2dec96c7fa5dadfbc2c","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":79.80000000000001,"rawValue":"0.798","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=9;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"ForgeCode","subject_type":"system"},"harnessId":"ForgeCode","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-65f07132a3fc7071ff4013f2","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":66.9,"rawValue":"0.669","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=30;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Crux","subject_type":"system"},"harnessId":"Crux","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10},{"id":"pub-5717243c51eb715f70af7e5c","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":75.3,"rawValue":"0.753","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=17;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Capy","subject_type":"system"},"harnessId":"Capy","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":11},{"id":"pub-4736554417b70ea13ceab7bf","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":76.4,"rawValue":"0.7640000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=14;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Meta-Harness","subject_type":"system"},"harnessId":"Meta-Harness","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":12},{"id":"pub-e829b4b6d85e9c8b25544ac7","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-04-16","metricId":"Accuracy mean","value":80.2,"rawValue":"0.802","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=8;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"WOZCODE","subject_type":"system"},"harnessId":"WOZCODE","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d5e55700f4450f7ab4746cee","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":42.8,"rawValue":"0.428","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=94;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-d561eba1e078e80796a228d5","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":40.1,"rawValue":"0.401","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=106;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-ac15a741396109eebdda7142","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":42.5,"rawValue":"0.425","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=101;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-2a4c758bad5979d4721f8da7","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":42.6,"rawValue":"0.426","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=100;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-e99015ad7fac067d4c5684cb","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":42.6,"rawValue":"0.42600000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=99;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a42ed10bbd1e045add6d47b1","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":42.7,"rawValue":"0.42700000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=96;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"MAYA","subject_type":"system"},"harnessId":"MAYA","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-695d45d5f53c819af0f30de4","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":46.5,"rawValue":"0.465","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=81;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"CAMEL-AI","subject_type":"system"},"harnessId":"CAMEL-AI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-63fc8f204e6649f8c2fdd2b0","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":43.10000000000001,"rawValue":"0.43100000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=91;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Goose","subject_type":"system"},"harnessId":"Goose","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-5a5ac54e224a83a7291fbf2c","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":40.1,"rawValue":"0.401","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=107;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Code","subject_type":"system"},"harnessId":"Claude Code","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-373419f17758115a28269f4c","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":42.8,"rawValue":"0.428","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=95;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-1849983e1d188d8b700e2677","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":42.5,"rawValue":"0.425","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=102;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":11},{"id":"pub-0bbb671a6eec61d21c8a4274","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy mean","value":42.7,"rawValue":"0.42700000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=98;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"MAYA-V2","subject_type":"system"},"harnessId":"MAYA-V2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":12},{"id":"pub-ad3d7110eb24f6d658a4779f","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-17","metricId":"Accuracy mean","value":53.400000000000006,"rawValue":"0.534","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=68;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Simplai Agent","subject_type":"system"},"harnessId":"Simplai Agent","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5be8b671057f2c161de40761","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-17","metricId":"Accuracy mean","value":53.400000000000006,"rawValue":"0.534","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=67;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Simplai Agent","subject_type":"system"},"harnessId":"Simplai Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-18ca020f217fb7a852e88e9c","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek/deepseek-v3.2","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-01","metricId":"Accuracy mean","value":39.6,"rawValue":"0.396","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=108;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-10","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f330ce216d52360ddd29ee2c","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":16.4,"rawValue":"0.16399999999999998","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=184;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d05e3fb4919a267d5f163ea7","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":17.1,"rawValue":"0.171","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=178;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-874ac16a60a169a4b3cb4b17","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":16.9,"rawValue":"0.16899999999999998","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=180;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-4deaa0470368d110c5a00fe0","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":15.4,"rawValue":"0.154","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=186;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Gemini CLI","subject_type":"system"},"harnessId":"Gemini CLI","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-cbd1853ec0a39305748f1e02","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":32.6,"rawValue":"0.326","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=131;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-a62274d5c4b7213147f60525","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":26.1,"rawValue":"0.261","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=154;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-9fa54e048636aae781460666","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":16.4,"rawValue":"0.16399999999999998","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=183;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-9a5ccdccd1ea35e25c791153","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":26.1,"rawValue":"0.261","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=153;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-6647c3ccc3ede3217381144b","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":32.6,"rawValue":"0.326","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=132;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":5},{"id":"pub-4d34337cdca371365319da01","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":19.6,"rawValue":"0.196","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=174;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Gemini CLI","subject_type":"system"},"harnessId":"Gemini CLI","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":6},{"id":"pub-4b0d97656771cb1b2dbb09c9","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":16.400000000000002,"rawValue":"0.164","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=181;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":7,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-11172d4c06fca6f798839036","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-06-17","metricId":"Accuracy mean","value":19.6,"rawValue":"0.196","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=173;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Gemini CLI","subject_type":"system"},"harnessId":"Gemini CLI","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":8,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-a1298b1c5d967cc767587b0c","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-19","metricId":"Accuracy mean","value":74.8,"rawValue":"0.748","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=19;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus-KIRA","subject_type":"system"},"harnessId":"Terminus-KIRA","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-92b82ae19a984604514efe22","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-19","metricId":"Accuracy mean","value":80.2,"rawValue":"0.802","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=7;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"TongAgents","subject_type":"system"},"harnessId":"TongAgents","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-5c766476f514b05d28686684","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-19","metricId":"Accuracy mean","value":78.4,"rawValue":"0.784","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=11;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"ForgeCode","subject_type":"system"},"harnessId":"ForgeCode","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-3bb33446990f12e39cdd6552","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-19","metricId":"Accuracy mean","value":78.4,"rawValue":"0.784","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=10;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Forge Code","subject_type":"system"},"harnessId":"Forge Code","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-06a50de460bb715d9c4ff562","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-19","metricId":"Accuracy mean","value":59.4,"rawValue":"0.594","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=52;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Gemini CLI","subject_type":"system"},"harnessId":"Gemini CLI","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-71c2a9f3ea41723acab69fbe","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-22","metricId":"Accuracy mean","value":33.4,"rawValue":"0.334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=129;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4812140fa98deff916a16246","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-22","metricId":"Accuracy mean","value":33.3,"rawValue":"0.33299999999999996","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=130;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Crux","subject_type":"system"},"harnessId":"Crux","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-a3aa5f42838ad454bbc38202","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-11","metricId":"Accuracy mean","value":52.400000000000006,"rawValue":"0.524","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=69;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f5790316f41b674f95505f1d","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":33.900000000000006,"rawValue":"0.339","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=127;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-9ba75b776aba86feb0fc2476","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":35.199999999999996,"rawValue":"0.352","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=120;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":5,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-65af4d28d75eedf5253cf19d","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":43.8,"rawValue":"0.438","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=85;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":7,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-031ea0ec0c11ffdaf834b509","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":49.6,"rawValue":"0.496","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=76;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":8,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-ed7444dcafd3720cd41bffcf","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":49.60000000000001,"rawValue":"0.49600000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=75;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b11dfb691b5e867fb9bd9e85","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":33.9,"rawValue":"0.33899999999999997","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=128;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-a7b838da888b5a4a7e9c50e0","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":43.8,"rawValue":"0.438","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=86;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-7b88b7363ade5f727e3cd6fe","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":35.2,"rawValue":"0.35200000000000004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=119;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-e141cd5ab208b7a12cf14c9d","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":29.2,"rawValue":"0.292","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=139;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-bd6e533c7f0b0726616b09a2","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":22.2,"rawValue":"0.222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=170;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-a4b76b1c3f76e64f185d0e6f","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":31.900000000000002,"rawValue":"0.319","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=133;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-481784c8c02fd50c86e00046","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":24.0,"rawValue":"0.24","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=163;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-cebcd6e008ffbd619a1cf302","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":22.2,"rawValue":"0.222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=171;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-4cd604993013777c401f4402","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":34.8,"rawValue":"0.348","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=125;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"spoox-m","subject_type":"system"},"harnessId":"spoox-m","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-462f5768a552a50d5ecf25f8","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":29.2,"rawValue":"0.292","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=141;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-45f6f732cb647a26ed9be2e7","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":34.8,"rawValue":"0.348","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=126;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"spoox-o-m","subject_type":"system"},"harnessId":"spoox-o-m","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-37126222e42a30618f649896","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":24.0,"rawValue":"0.24","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=164;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-2647775565f8c95b4a9a5cf1","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":31.900000000000002,"rawValue":"0.319","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=134;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10},{"id":"pub-a16ac277ea727da5466001da","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":7.9,"rawValue":"0.079","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=198;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-91bc8bc0ed283fbe8bd48991","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":11.5,"rawValue":"0.115","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=193;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-3a5428aed69da8f431abd92d","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":9.9,"rawValue":"0.099","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=195;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-0d85443c218e676cebdab475","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":7.000000000000001,"rawValue":"0.07","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=200;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-fe57af4e7a74dd2350318f14","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":21.8,"rawValue":"0.218","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=172;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"spoox-o-m","subject_type":"system"},"harnessId":"spoox-o-m","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fbaa4077ea0403a8c8f963f6","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":9.9,"rawValue":"0.099","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=196;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e4c98b2f361c7b0a4603322f","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":7.000000000000001,"rawValue":"0.07","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=201;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-03","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mini-SWE-Agent","subject_type":"system"},"harnessId":"Mini-SWE-Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-daba28d634432d8efcd831f5","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":7.9,"rawValue":"0.079","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=199;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-c3718e7e69964d2376b2fcd9","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy mean","value":11.5,"rawValue":"0.115","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=194;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-de3c6a0cd9f4b592c81949f2","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-13","metricId":"Accuracy mean","value":47.599999999999994,"rawValue":"0.476","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=78;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-21e90f53b8e6269b9df981ad","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-13","metricId":"Accuracy mean","value":47.599999999999994,"rawValue":"0.476","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=79;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-4b50a3ac57f019e7da67e308","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-13","metricId":"Accuracy mean","value":47.6,"rawValue":"0.47600000000000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=77;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ff832f99704c74fcb6c5b5c9","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"gpt-5.1-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-12","metricId":"Accuracy mean","value":53.5,"rawValue":"0.535","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=66;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Letta Code","subject_type":"system"},"harnessId":"Letta Code","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-2e185f0c0231e98d7a6dc2d8","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"gpt-5.1-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-12","metricId":"Accuracy mean","value":57.8,"rawValue":"0.578","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=58;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Crux","subject_type":"system"},"harnessId":"Crux","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-223aaebafd05824e3c2ffcc2","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"gpt-5.1-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-12","metricId":"Accuracy mean","value":57.8,"rawValue":"0.578","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=57;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-1f669adfb416145c4296987b","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"gpt-5.1-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-12","metricId":"Accuracy mean","value":36.9,"rawValue":"0.369","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=114;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-062cbe51489bf5c754891c03","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"gpt-5.1-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-12","metricId":"Accuracy mean","value":36.9,"rawValue":"0.369","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=111;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-b67a257de5e068cb99b3d100","canonicalModelId":"openai/gpt-5.1-codex-max@2025-11-19","modelName":"GPT-5.1 Codex Max","modelRef":"gpt-5.1-codex-max","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-19","metricId":"Accuracy mean","value":60.4,"rawValue":"0.604","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=51;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-9d135192f711eadafaa0d550","canonicalModelId":"openai/gpt-5.1-codex-mini@2025-11-17","modelName":"GPT-5.1 Codex Mini","modelRef":"gpt-5.1-codex-mini","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-12","metricId":"Accuracy mean","value":43.1,"rawValue":"0.431","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=92;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-9432acd6f93f494a891422a7","canonicalModelId":"openai/gpt-5.1-codex-mini@2025-11-17","modelName":"GPT-5.1 Codex Mini","modelRef":"gpt-5.1-codex-mini","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-12","metricId":"Accuracy mean","value":43.10000000000001,"rawValue":"0.43100000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=90;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Crux","subject_type":"system"},"harnessId":"Crux","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-4c790dea209bc149bcc4122b","canonicalModelId":"openai/gpt-5.1-codex-mini@2025-11-17","modelName":"GPT-5.1 Codex Mini","modelRef":"gpt-5.1-codex-mini","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-11-12","metricId":"Accuracy mean","value":61.6,"rawValue":"0.616","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=47;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"hookele","subject_type":"system"},"harnessId":"hookele","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-384b5625c83ba60067a30e76","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy mean","value":64.9,"rawValue":"0.649","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=35;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Droid","subject_type":"system"},"harnessId":"Droid","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-188516ff76d0cf439e9965dc","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy mean","value":62.9,"rawValue":"0.629","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=41;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-ac8604a0bd96988b3ede2789","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy mean","value":62.9,"rawValue":"0.629","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=42;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8681364fa8e94651ec469bad","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy mean","value":60.70000000000001,"rawValue":"0.6070000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=50;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mux","subject_type":"system"},"harnessId":"Mux","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-7fd5dce78d388362ec6c86da","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy mean","value":64.9,"rawValue":"0.649","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=36;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Droid","subject_type":"system"},"harnessId":"Droid","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-2cb54dcab956c9d7a217d7e6","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy mean","value":54.0,"rawValue":"0.54","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=65;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-3d0080f5d43551bed04eeaae","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2025-12-18","metricId":"Accuracy mean","value":66.5,"rawValue":"0.665","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=32;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Deep Agents","subject_type":"system"},"harnessId":"Deep Agents","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f4289c88f2d1992eb81c4ba5","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":75.1,"rawValue":"0.751","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=18;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-06","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Simple Codex","subject_type":"system"},"harnessId":"Simple Codex","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-f1dcccf11e75d8fbd7e6bc62","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":71.5,"rawValue":"0.715","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=24;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"spoox-o-m","subject_type":"system"},"harnessId":"spoox-o-m","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e0d7e1522555df2a44b3bcc3","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":75.7,"rawValue":"0.757","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=16;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codelia","subject_type":"system"},"harnessId":"Codelia","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-df56a45d43539b9a1addadda","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":77.3,"rawValue":"0.773","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=13;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Droid","subject_type":"system"},"harnessId":"Droid","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-d639a48710811f2d9cf64641","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":70.3,"rawValue":"0.703","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=25;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-10","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"CodeBrain-1","subject_type":"system"},"harnessId":"CodeBrain-1","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-c3a5662d7580a6c949716a97","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":64.7,"rawValue":"0.647","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=37;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":6},{"id":"pub-baa7428664884ab66ddf959d","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":74.6,"rawValue":"0.746","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=21;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-06","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Mux","subject_type":"system"},"harnessId":"Mux","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":7},{"id":"pub-b91597b34cf519e7af844991","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":78.4,"rawValue":"0.784","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=12;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"SageAgent","subject_type":"system"},"harnessId":"SageAgent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":8},{"id":"pub-3d8eefb9abd3fcb3f76b683e","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":75.8,"rawValue":"0.758","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=15;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-10","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"CodeBrain-1.5","subject_type":"system"},"harnessId":"CodeBrain-1.5","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":9},{"id":"pub-1d8508f36dbc42076b2eb714","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy mean","value":69.1,"rawValue":"0.691","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=29;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"IndusAGI Coding Agent","subject_type":"system"},"harnessId":"IndusAGI Coding Agent","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":10},{"id":"pub-0d9084acee734cd4bb5e8400","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-03-05","metricId":"Accuracy mean","value":81.8,"rawValue":"0.818","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=6;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"ForgeCode","subject_type":"system"},"harnessId":"ForgeCode","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e36d5144b0bab71e905a8e8a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy mean","value":83.1,"rawValue":"0.831","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=3;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Capy","subject_type":"system"},"harnessId":"Capy","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b760ca45b06556c138fd4cb1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy mean","value":82.2,"rawValue":"0.8220000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=4;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex CLI","subject_type":"system"},"harnessId":"Codex CLI","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-655816fcab5ef07789cb4192","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy mean","value":66.1,"rawValue":"0.6609999999999999","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=33;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"clnkr","subject_type":"system"},"harnessId":"clnkr","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-55b7fae8661724a2f2001e9d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy mean","value":82.0,"rawValue":"0.8200000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=5;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Codex","subject_type":"system"},"harnessId":"Codex","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-0d475b1bec6ec1c307db522c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy mean","value":84.7,"rawValue":"0.8470000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=2;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"NexAU-AHE","subject_type":"system"},"harnessId":"NexAU-AHE","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-2ec4f525feeb31b314baf29f","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-01-27","metricId":"Accuracy mean","value":43.2,"rawValue":"0.43200000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=89;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-04","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Terminus 2","subject_type":"system"},"harnessId":"Terminus 2","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5075b74fb1afdd316a71385b","canonicalModelId":"qwen/qwen3.5-9b@current","modelName":"Qwen3.5 9B","modelRef":"qwen3.5-9B","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-02-24","metricId":"Accuracy mean","value":9.2,"rawValue":"0.092","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=197;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"little-coder","subject_type":"system"},"harnessId":"little-coder","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5f298746fb7cb46820d5df3b","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b","benchmarkId":"epoch-terminalbench_external","benchmarkVersion":"2026-04-14","metricId":"Accuracy mean","value":23.0,"rawValue":"0.23","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=169;column=Accuracy mean","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"little-coder","subject_type":"system"},"harnessId":"little-coder","qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3c3f24d5c5cf47ad8a91385d","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp","benchmarkId":"epoch-the_agent_company_external","benchmarkVersion":"2025-09-29","metricId":"% Score","value":52.400000000000006,"rawValue":"0.524","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"the_agent_company_external.csv:row=13;column=% Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-48c103db4283bf7e651a77c1","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2025-09-29","metricId":"Score","value":1034.0040000000001,"rawValue":"1034.0040000000001","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=46;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-df00b9c3a14d094e59929852","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-02-12","metricId":"Score","value":-23.15999999999999,"rawValue":"-23.15999999999999","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=59;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c3dc6db12d3ed61f0b39b33c","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-06-01","metricId":"Score","value":2157.7716666666665,"rawValue":"2157.7716666666665","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=42;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c69dcb8f4d572bed9d7544ed","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":4529.944000000001,"rawValue":"4529.944000000001","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=29;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-15c18a98b076327348ac78ec","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":5680.258,"rawValue":"5680.258","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=18;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":5},{"id":"pub-d76ac35cd74738b40462b8a1","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_low","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":5018.519999999999,"rawValue":"5018.519999999999","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=25;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-c64b09e9686886c6dd4b9b74","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":4966.639999999998,"rawValue":"4966.639999999998","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=27;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-fea4823425165dc753642c6d","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_medium","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-06-09","metricId":"Score","value":4339.812500000001,"rawValue":"4339.812500000001","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=31;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a45d84b0b13b73e3d0dc117f","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2025-10-15","metricId":"Score","value":458.88599999999985,"rawValue":"458.88599999999985","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=51;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5c0952d20fa2b6e95c8d60c9","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2025-11-24","metricId":"Score","value":4967.056666666668,"rawValue":"4967.056666666668","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=26;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f2b48f7aa60d88993523517f","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-02-05","metricId":"Score","value":8017.594,"rawValue":"8017.594","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=8;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-63d714cbb90889eb8236afc6","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-04-16","metricId":"Score","value":10936.763333333334,"rawValue":"10936.763333333334","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=3;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-413ce34198252a034be8a2b7","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":2992.344,"rawValue":"2992.344","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=40;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-911712ecdb3689b33266c532","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-05-28","metricId":"Score","value":5787.428,"rawValue":"5787.428","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=17;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e1fefd93818186c330feb63b","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-07-24","metricId":"Score","value":11181.87333333333,"rawValue":"11181.87333333333","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=2;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-07dddd746fd44147f0a895bc","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2025-09-29","metricId":"Score","value":3838.7400000000002,"rawValue":"3838.7400000000002","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=35;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-82e6ec121c2ea45b6cc39c1c","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-02-17","metricId":"Score","value":7204.139999999999,"rawValue":"7204.139999999999","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=11;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c764cd8fe4dd09783c5f1449","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-06-30","metricId":"Score","value":6377.701666666665,"rawValue":"6377.701666666665","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=13;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6db72572f0637803753a965b","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-04-24","metricId":"Score","value":3284.519999999999,"rawValue":"3284.519999999999","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=39;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-afe0d9081c13e2fb7e2f57c8","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2025-06-17","metricId":"Score","value":548.836,"rawValue":"548.836","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=49;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5d3efbe88361aaca994051ea","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2025-06-17","metricId":"Score","value":573.6420000000003,"rawValue":"573.6420000000003","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=48;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f534c925bfad82f154570182","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-02-19","metricId":"Score","value":911.2139999999997,"rawValue":"911.2139999999997","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=47;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-347378eb0350356905112977","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-05-19","metricId":"Score","value":5396.418333333334,"rawValue":"5396.418333333334","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=21;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a2137ebc007e8e973606b47f","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2025-12-22","metricId":"Score","value":2376.816,"rawValue":"2376.816","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=41;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5e3462b84705f62ed1eb8c8c","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-02-11","metricId":"Score","value":4432.122000000001,"rawValue":"4432.122000000001","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=30;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1911749fb90020c58766e95d","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-04-07","metricId":"Score","value":5634.406,"rawValue":"5634.406","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=19;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e733952299d5e89dc23b9a4d","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-06-16","metricId":"Score","value":8313.778333333335,"rawValue":"8313.778333333335","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=6;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7b7534495c6de64032959244","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-08-14","metricId":"Score","value":8163.609999999998,"rawValue":"8163.609999999998","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=7;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a1e28f66fd9963d38d4fc424","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2025-08-07","metricId":"Score","value":-31.18399999999995,"rawValue":"-31.18399999999995","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=60;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8a8324610e0b7576cc6126cc","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2025-11-13","metricId":"Score","value":1473.432,"rawValue":"1473.432","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=43;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9f7fe6a7da6e9ad26c2d1e7a","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2025-12-11","metricId":"Score","value":3591.3259999999996,"rawValue":"3591.3259999999996","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=38;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-43eb37a8c07a4f8366869a0d","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-02-05","metricId":"Score","value":5940.124000000002,"rawValue":"5940.124000000002","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=16;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-63893b9efe1bfacbeb418364","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-03-05","metricId":"Score","value":6144.176,"rawValue":"6144.176","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=15;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-648b2b4e6f4cb5797222dc3a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-04-23","metricId":"Score","value":7523.843999999999,"rawValue":"7523.843999999999","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=9;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-21a4eb23142f1d6041cf9c29","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":4094.7120000000004,"rawValue":"4094.7120000000004","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=33;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cdcfa46a82adc41853d2f906","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":9619.367999999999,"rawValue":"9619.367999999999","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=4;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ce6523ac8ad811bd9f4465e1","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-07-09","metricId":"Score","value":7343.206,"rawValue":"7343.206","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=10;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1e46ddf23e0ea3d4b389d862","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-07-08","metricId":"Score","value":3887.4259999999986,"rawValue":"3887.4259999999986","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=34;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-16ba2d73b562756428d521df","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-08-12","metricId":"Score","value":9047.032500000001,"rawValue":"9047.032500000001","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=5;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1b54e809f3ea3abc56f17cf1","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-01-27","metricId":"Score","value":1198.4579999999999,"rawValue":"1198.4579999999999","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=44;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-88eb81cb96d83a12475f664d","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-04-20","metricId":"Score","value":6204.570000000001,"rawValue":"6204.570000000001","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=14;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-bfab718ca974d8c592941def","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-06-12","metricId":"Score","value":5082.936,"rawValue":"5082.936","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=24;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0912338dfab9b9cb0075765b","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_unknown","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-07-16","metricId":"Score","value":5165.035000000001,"rawValue":"5165.035000000001","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=22;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-25170f3786f1b93186e48d2e","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27B","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-02-24","metricId":"Score","value":201.97599999999994,"rawValue":"201.97599999999994","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=52;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f14d2ec0b0780dbeeecac235","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-02-25","metricId":"Score","value":462.68500000000023,"rawValue":"462.68500000000023","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=50;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0d0fdbdd7eb4ae0accb40ff5","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-02-16","metricId":"Score","value":0.5440000000000417,"rawValue":"0.5440000000000417","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=56;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-794b8fd3cedf6803b7b17970","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-04-20","metricId":"Score","value":4254.1866666666665,"rawValue":"4254.1866666666665","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=32;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2027dee135d9ed0ef608baf2","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"epoch-vending_bench_2_external","benchmarkVersion":"2026-03-31","metricId":"Score","value":5114.872000000001,"rawValue":"5114.872000000001","unit":"score","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vending_bench_2_external.csv:row=23;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_unverified"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-27e68d495d256d085102769c","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-05-22","metricId":"Correct","value":33.0,"rawValue":"0.33","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=35;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-00ec2589ea0a236cacf8d8ad","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_16K","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-05-22","metricId":"Correct","value":38.0,"rawValue":"0.38","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=25;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-50c6439828295849dfad3d8b","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_32K","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-11-24","metricId":"Correct","value":40.0,"rawValue":"0.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=20;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://x.com/ChaseBrowe32432/status/1993038782975951149"},{"id":"pub-4c4a505b520f2b03c6160ba8","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-09-29","metricId":"Correct","value":38.0,"rawValue":"0.38","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=26;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-22e68fc2435c44c601473252","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-09-29","metricId":"Correct","value":39.800000000000004,"rawValue":"0.398","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=21;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-cde7d01f05852e28356e780c","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-08-07","metricId":"Correct","value":66.0,"rawValue":"0.66","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=6;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-0424cae00f95fe7e07a345af","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-08-07","metricId":"Correct","value":63.2,"rawValue":"0.632","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=7;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-bc7d22f4916868c21eb86d2a","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-08-07","metricId":"Correct","value":39.0,"rawValue":"0.39","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=23;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-ecd66aa0ba805a074db62a92","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-08-07","metricId":"Correct","value":40.2,"rawValue":"0.402","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=18;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-ef4449402efebe1839d827ee","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-08-07","metricId":"Correct","value":37.2,"rawValue":"0.372","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=27;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-3d6d7de09e08db84872e76ab","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-08-07","metricId":"Correct","value":35.4,"rawValue":"0.354","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=29;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-0288541cd8ba1db33aae5805","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-11-13","metricId":"Correct","value":58.699999999999996,"rawValue":"0.587","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=8;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-46506a86b6564996b98a141a","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-11-13","metricId":"Correct","value":53.300000000000004,"rawValue":"0.533","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=10;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-5856285de606a2196803dc20","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-12-11","metricId":"Correct","value":67.0,"rawValue":"0.67","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=5;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-f2d022b1b6b3a187950f6781","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-vpct_external","benchmarkVersion":"2025-12-11","metricId":"Correct","value":84.0,"rawValue":"0.84","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"vpct_external.csv:row=3;column=Correct","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://cbrower.dev/vpct"},{"id":"pub-8f3576e4a85792ef63a5f6ca","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-09-29","metricId":"Arena Score","value":1271.77,"rawValue":"1271.77","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=107;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-64b18a44cc1c7c8a411a078a","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp_thinking","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-09-29","metricId":"Arena Score","value":1367.0,"rawValue":"1367.0","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=86;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e22bc38c5cd245ea1edf89ac","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"MiniMax-M2.1","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-12-23","metricId":"Arena Score","value":1386.99,"rawValue":"1386.99","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=78;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-88233b798cf65ebd025bb4f3","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax-M2.5","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-02-12","metricId":"Arena Score","value":1383.9,"rawValue":"1383.9","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=81;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ce2316dc3024bd677134747b","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMax-M2.7","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-03-18","metricId":"Arena Score","value":1398.14,"rawValue":"1398.14","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=71;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-55bb03ba7ffc1ac53c613e56","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-06-01","metricId":"Arena Score","value":1487.64,"rawValue":"1487.64","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=42;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2ab8ed62d0405eb95de75ab2","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-07-31","metricId":"Arena Score","value":1273.25,"rawValue":"1273.25","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=106;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-66093b507d7e2457817d5fd7","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-06-09","metricId":"Arena Score","value":1626.26,"rawValue":"1626.26","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=7;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4a6e6ae129ec78fe17aa6b10","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-10-15","metricId":"Arena Score","value":1286.0,"rawValue":"1286.0","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=105;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-53fbf1742bc08dbf9aa5b66f","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-10-15","metricId":"Arena Score","value":1327.41,"rawValue":"1327.41","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=100;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2a0e9171a5886b01a8a53ebd","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-11-24","metricId":"Arena Score","value":1468.08,"rawValue":"1468.08","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=48;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-86e76c854e9a7e7ca0ecaaff","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_32K","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-11-24","metricId":"Arena Score","value":1494.39,"rawValue":"1494.39","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=40;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-80cd5c8041888031d2ee35cb","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-02-05","metricId":"Arena Score","value":1536.27,"rawValue":"1536.27","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=27;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4e9be1a187d0ff284da1d22e","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-02-05","metricId":"Arena Score","value":1545.68,"rawValue":"1545.68","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=22;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-2db44023330516536d5864f9","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-02-05","metricId":"Arena Score","value":1556.29,"rawValue":"1556.29","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=20;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-01dc81c524664e3637f41fdd","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-16","metricId":"Arena Score","value":1557.52,"rawValue":"1557.52","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=17;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-fb6b122af31d5cd417be90f0","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-16","metricId":"Arena Score","value":1556.7,"rawValue":"1556.7","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=18;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bd02722db2912d099fb1477a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-16","metricId":"Arena Score","value":1566.85,"rawValue":"1566.85","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=15;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d6976388f8758a75f5738658","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-05-28","metricId":"Arena Score","value":1563.49,"rawValue":"1563.49","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=16;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8f00141e8e8d1516b8ccd7b4","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_none","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-05-28","metricId":"Arena Score","value":1539.1,"rawValue":"1539.1","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=23;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-c826f3330d184f45995679a7","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-05-28","metricId":"Arena Score","value":1552.22,"rawValue":"1552.22","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=21;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-77a546a442d6f55538df6971","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-07-24","metricId":"Arena Score","value":1662.59,"rawValue":"1662.59","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=5;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-35932c1b56ae9221e273bac9","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-07-24","metricId":"Arena Score","value":1690.64,"rawValue":"1690.64","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=2;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-160525ab4acd0f1458dfab0c","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-09-29","metricId":"Arena Score","value":1385.51,"rawValue":"1385.51","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=80;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-b0a5137e9b28993a463a4170","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-09-29","metricId":"Arena Score","value":1391.0,"rawValue":"1391.0","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=75;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9bbdd67ca6c3e54c825f0715","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-09-29","metricId":"Arena Score","value":1353.0,"rawValue":"1353.0","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=92;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-0f044e22b554f42dcbf36fbb","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-02-17","metricId":"Arena Score","value":1522.47,"rawValue":"1522.47","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=30;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-684a7a955a0a86d8410a7212","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-06-30","metricId":"Arena Score","value":1538.58,"rawValue":"1538.58","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=26;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9abf145fcd16e3e380512a14","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-07-31","metricId":"Arena Score","value":1579.1,"rawValue":"1579.1","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=14;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-a9b32a80cfafbbdb96659349","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-24","metricId":"Arena Score","value":1431.07,"rawValue":"1431.07","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=64;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-88a1b0e4470e8f6779f2a96a","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-08-13","metricId":"Arena Score","value":1581.64,"rawValue":"1581.64","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=13;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a208f2c971d26c9194f03f62","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-24","metricId":"Arena Score","value":1463.8,"rawValue":"1463.8","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=49;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-fe244b2105db24f47c8f2e1b","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_none","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-24","metricId":"Arena Score","value":1445.46,"rawValue":"1445.46","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=56;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-abf5fa4eea2107aa7158633f","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-24","metricId":"Arena Score","value":1463.63,"rawValue":"1463.63","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=50;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-fa78cc7d41199e52174f9020","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek/deepseek-v3.2","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-12-01","metricId":"Arena Score","value":1324.54,"rawValue":"1324.54","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=101;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e96dabf103a2ee1541103fb5","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-06-17","metricId":"Arena Score","value":1225.69,"rawValue":"1225.69","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=117;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-790c8e208632eb20b088bec1","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-02-19","metricId":"Arena Score","value":1446.46,"rawValue":"1446.46","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=55;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-83ea72ef77f83e0454942c86","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-07-21","metricId":"Arena Score","value":1449.14,"rawValue":"1449.14","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=54;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-85c95df727d7549c8783580b","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-05-19","metricId":"Arena Score","value":1498.59,"rawValue":"1498.59","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=39;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e1e80f26300163cf04e342be","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_medium","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-05-19","metricId":"Arena Score","value":1489.74,"rawValue":"1489.74","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=41;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7583829ac8634940ebc561d4","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-05-19","metricId":"Arena Score","value":1506.38,"rawValue":"1506.38","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=38;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-235fe21b00748fb4dcfad298","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-07-21","metricId":"Arena Score","value":1538.91,"rawValue":"1538.91","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=25;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-fc49cfbc1b1589a8dcaa8a94","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-07-21","metricId":"Arena Score","value":1527.83,"rawValue":"1527.83","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=29;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3c5b56475faea4e56f3c77e6","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-08-13","metricId":"Arena Score","value":1587.04,"rawValue":"1587.04","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=12;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2e3e8b3f6d615dc5bff47f5c","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-12-22","metricId":"Arena Score","value":1434.29,"rawValue":"1434.29","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=62;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a8d1ed720d6ec24ebd66e122","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-02-11","metricId":"Arena Score","value":1435.46,"rawValue":"1435.46","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=61;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0583c049080a1e2e404c59b6","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-07","metricId":"Arena Score","value":1508.93,"rawValue":"1508.93","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=35;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9ff2cce9a9e198157e612975","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-06-16","metricId":"Arena Score","value":1593.25,"rawValue":"1593.25","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=11;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bcb9a5cdaaf2cdcdcabee4f8","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3_max","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-08-14","metricId":"Arena Score","value":1598.6,"rawValue":"1598.6","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=9;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fa5c7d5da3a102e05b064345","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-08-07","metricId":"Arena Score","value":1419.14,"rawValue":"1419.14","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=65;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7c954529b472de07ec2b3485","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_low","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-11-13","metricId":"Arena Score","value":1359.0,"rawValue":"1359.0","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=89;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-472c1914c29f547ceb40d898","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-11-13","metricId":"Arena Score","value":1391.01,"rawValue":"1391.01","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=74;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ec0096d5dcb6e7c8c241210e","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-11-13","metricId":"Arena Score","value":1340.63,"rawValue":"1340.63","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=94;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-de5641f0fba652254f991676","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"gpt-5.1-codex","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-11-12","metricId":"Arena Score","value":1336.28,"rawValue":"1336.28","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=98;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-208e79e7dea20ce65cf7de0e","canonicalModelId":"openai/gpt-5.1-codex-mini@2025-11-17","modelName":"GPT-5.1 Codex Mini","modelRef":"gpt-5.1-codex-mini","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-11-12","metricId":"Arena Score","value":1247.0,"rawValue":"1247.0","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=112;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7a6b384dd2c50015296582e2","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-12-11","metricId":"Arena Score","value":1480.0,"rawValue":"1480.0","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=44;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c3d2874c9cb26e93d7968fcd","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-12-11","metricId":"Arena Score","value":1394.0,"rawValue":"1394.0","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=73;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2482631cc1eb18cdd64d827f","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-12-11","metricId":"Arena Score","value":1417.81,"rawValue":"1417.81","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=66;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-771e474c3e78b0de496215ae","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-12-18","metricId":"Arena Score","value":1338.28,"rawValue":"1338.28","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=96;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d43a1fde529a07e948835c4b","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-02-05","metricId":"Arena Score","value":1369.0,"rawValue":"1369.0","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=85;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ac350eff854823115f1a9e4a","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-03-05","metricId":"Arena Score","value":1462.9,"rawValue":"1462.9","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=51;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-e6976acafa1e7d0061a4b0b1","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_medium","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-03-05","metricId":"Arena Score","value":1442.16,"rawValue":"1442.16","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=57;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0c03b913abe5a362d9f44935","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-03-05","metricId":"Arena Score","value":1390.56,"rawValue":"1390.56","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=76;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-811a0cbf7ac324f979ac8d49","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-03-17","metricId":"Arena Score","value":1397.08,"rawValue":"1397.08","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=72;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-105df3847889560843ef96f6","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-23","metricId":"Arena Score","value":1486.17,"rawValue":"1486.17","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=43;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-575bb82b8abd5aa7d5bfef4e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-23","metricId":"Arena Score","value":1457.51,"rawValue":"1457.51","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=53;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2c32ebf4d3b56221f1623b9a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-23","metricId":"Arena Score","value":1508.35,"rawValue":"1508.35","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=37;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-7759c88c0476bb50fcf877e4","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_xhigh","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-07-09","metricId":"Arena Score","value":1517.6,"rawValue":"1517.6","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=32;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3813200b239a7e973640cf87","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_xhigh","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-07-09","metricId":"Arena Score","value":1618.71,"rawValue":"1618.71","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=8;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e0502b213415f5492c56fe8f","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_xhigh","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-07-09","metricId":"Arena Score","value":1519.87,"rawValue":"1519.87","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=31;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2609d18b4db1bf8b2382620d","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2025-09-19","metricId":"Arena Score","value":1149.0,"rawValue":"1149.0","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=123;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-dbefc6281965d6d316ab145a","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-07-08","metricId":"Arena Score","value":1556.36,"rawValue":"1556.36","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=19;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-88cb4c008ca1ae3b1c521757","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_high","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-08-12","metricId":"Arena Score","value":1628.84,"rawValue":"1628.84","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=6;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c652cfeb6cc920c7f9a2a61b","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-01-27","metricId":"Arena Score","value":1435.92,"rawValue":"1435.92","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=60;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6136a717a0080bd960e75a23","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5_none","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-01-27","metricId":"Arena Score","value":1404.86,"rawValue":"1404.86","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=68;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-dba80fd728c27f4ce2fa4351","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-20","metricId":"Arena Score","value":1508.71,"rawValue":"1508.71","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=36;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5c96ff3346d130abd1e5bc64","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-06-12","metricId":"Arena Score","value":1472.65,"rawValue":"1472.65","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=47;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-54040bf5cc5f335e9f6d509e","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-07-16","metricId":"Arena Score","value":1673.7,"rawValue":"1673.7","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=3;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["heuristic_model_mapping","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-30b7f1c70d2b51e53d2ef553","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"mimo-v2-flash","benchmarkId":"epoch-webdev_arena_external","metricId":"Arena Score","value":1330.17,"rawValue":"1330.17","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=99;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-662649f11b6a639f490e5e2a","canonicalModelId":"xiaomi/mimo-v2-pro@current","modelName":"MiMo V2 Pro","modelRef":"mimo-v2-pro","benchmarkId":"epoch-webdev_arena_external","metricId":"Arena Score","value":1433.51,"rawValue":"1433.51","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=63;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1e727b380cef7aaab1e87b4a","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"mimo-v2.5","benchmarkId":"epoch-webdev_arena_external","metricId":"Arena Score","value":1437.8,"rawValue":"1437.8","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=59;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-289943ecab5faa7729b6b8fe","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"mimo-v2.5-pro","benchmarkId":"epoch-webdev_arena_external","metricId":"Arena Score","value":1475.65,"rawValue":"1475.65","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=46;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6416c4bfaa89ec6ed12805b8","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27B","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-02-24","metricId":"Arena Score","value":1357.46,"rawValue":"1357.46","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=90;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4616cb178134e0757133c488","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-02-25","metricId":"Arena Score","value":1238.12,"rawValue":"1238.12","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=114;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ac1555a08ca310a7608f8cc9","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-02-16","metricId":"Arena Score","value":1399.23,"rawValue":"1399.23","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=70;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-56abe9e2e9fb4145bf3fba03","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-14","metricId":"Arena Score","value":1249.75,"rawValue":"1249.75","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=111;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-673da282ad4290da08da3986","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-04-20","metricId":"Arena Score","value":1478.88,"rawValue":"1478.88","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=45;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8e2f84e67339ce03369de046","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-03-31","metricId":"Arena Score","value":1459.74,"rawValue":"1459.74","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=52;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8b929557929701e15b32aea0","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-05-19","metricId":"Arena Score","value":1516.65,"rawValue":"1516.65","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=33;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9ce067bc33b726849b4262c8","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-08-14","metricId":"Arena Score","value":1594.78,"rawValue":"1594.78","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=10;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-494da694f7fb89e7435daf28","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max_unknown","benchmarkId":"epoch-webdev_arena_external","benchmarkVersion":"2026-08-02","metricId":"Arena Score","value":1668.94,"rawValue":"1668.94","unit":"elo","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"webdev_arena_external.csv:row=4;column=Arena Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","source_version_label_without_id","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-70a51f6f6461f3e5d94946d8","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-05-28","metricId":"Accuracy","value":41.63,"rawValue":"0.41630000000000006","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=100;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-273b8cf64b7eb73fcc09af3b","canonicalModelId":"deepseek/v3.1@2025-08-21","modelName":"DeepSeek V3.1","modelRef":"DeepSeek-V3.1","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-08-21","metricId":"Accuracy","value":37.5,"rawValue":"0.375","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=124;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-949888bd402c6dedab3ae8ab","canonicalModelId":"deepseek/v3.1@2025-08-21","modelName":"DeepSeek V3.1","modelRef":"DeepSeek-V3.1_thinking","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-08-21","metricId":"Accuracy","value":38.36,"rawValue":"0.3836","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=118;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-30aeea61146e62cf278af6d1","canonicalModelId":"deepseek/v3.2-exp@2025-09-29","modelName":"DeepSeek V3.2 Exp","modelRef":"DeepSeek-V3.2-Exp_thinking","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy","value":39.48,"rawValue":"0.3948","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=110;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-ea0616a8cf3375a6e169c9f4","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMax-M2.7","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-03-18","metricId":"Accuracy","value":36.95,"rawValue":"0.36950000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=128;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-18","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-97ba5dd6e240d541b2fd74c7","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-07-31","metricId":"Accuracy","value":41.17,"rawValue":"0.4117","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=101;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-74961a82b346bea89c0b39e5","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-06-09","metricId":"Accuracy","value":87.85,"rawValue":"0.8785","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=7;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0f969159ceef495b3d34ad7b","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-06-09","metricId":"Accuracy","value":91.94,"rawValue":"0.9194","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=2;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b24157f15b2ea85f16c1673e","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy","value":45.4,"rawValue":"0.454","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=85;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-26c6d56dc7e55f669ba0bf25","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_16K","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-10-15","metricId":"Accuracy","value":44.15,"rawValue":"0.4415","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=86;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-27ed8fd3c63abbb190f4e26d","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_16K","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-05-22","metricId":"Accuracy","value":43.72,"rawValue":"0.4372","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=89;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-cafe8ac867cdaac8e045ded9","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_16K","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-11-24","metricId":"Accuracy","value":63.74000000000001,"rawValue":"0.6374000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=37;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-a9cc464de77a77276dcd4bb0","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy","value":65.87,"rawValue":"0.6587000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=36;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-85f8b7d832cbc59c3dd4f93b","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy","value":77.95,"rawValue":"0.7795000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=16;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-0cc5de6376dbcfd17688bbd3","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_unknown","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy","value":77.9,"rawValue":"0.779","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=17;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-a4dc12ddf962149aa8e22af0","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-16","metricId":"Accuracy","value":76.4,"rawValue":"0.7640000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=21;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-77431d193ee3b61dcd73ae53","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-16","metricId":"Accuracy","value":76.44,"rawValue":"0.7644","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=20;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-8043bef1121db11ff2a3b4ac","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-16","metricId":"Accuracy","value":75.45,"rawValue":"0.7545000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=24;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-714a7aa316b1adfc8ad0d030","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-16","metricId":"Accuracy","value":76.4,"rawValue":"0.764","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=22;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-ed5a01ca6542e67a88d49c20","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_medium","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-05-28","metricId":"Accuracy","value":76.04,"rawValue":"0.7604000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=23;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-cbb6694a37c21c08de653f90","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_none","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-05-28","metricId":"Accuracy","value":70.45,"rawValue":"0.7045","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=27;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-fbdbb17c31e821b4ab1b3621","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_xhigh","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-05-28","metricId":"Accuracy","value":82.89,"rawValue":"0.8289","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=12;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d005807c0b03490f68e4cbf4","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-24","metricId":"Accuracy","value":86.34,"rawValue":"0.8634000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=9;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-80414d34671bd45745007156","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-24","metricId":"Accuracy","value":91.59,"rawValue":"0.9159","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=4;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-55e653ae540f253d4a4b15c3","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-24","metricId":"Accuracy","value":91.78,"rawValue":"0.9178000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=3;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-47db27bb054e48c7e4c8d03e","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy","value":46.7,"rawValue":"0.467","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=76;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-0d678187910210d9d8aa3449","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_16K","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-09-29","metricId":"Accuracy","value":47.71,"rawValue":"0.4771","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=73;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-28115bc5890fca3360d2207d","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_medium","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-02-17","metricId":"Accuracy","value":66.07,"rawValue":"0.6607","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=35;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-cda4a2b59b7a1562481db81b","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-06-30","metricId":"Accuracy","value":68.78,"rawValue":"0.6878000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=30;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-259de5a87e31a39eee0cee24","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-31","metricId":"Accuracy","value":57.05,"rawValue":"0.5705","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=52;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-292d44cf56c3a7a1f8a22b1f","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-31","metricId":"Accuracy","value":62.96000000000001,"rawValue":"0.6296","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=39;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-8186a8d774b191fbe1d4e7a0","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-24","metricId":"Accuracy","value":43.76,"rawValue":"0.4376","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=88;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-c2e06de300e56dca390192e5","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash_max","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-24","metricId":"Accuracy","value":45.63,"rawValue":"0.45630000000000004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=83;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e5ad11e3db1c1b66bd869c6b","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_max","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-08-13","metricId":"Accuracy","value":66.22,"rawValue":"0.6622","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=34;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a3ee02cc2c216d270f498c16","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-24","metricId":"Accuracy","value":46.53,"rawValue":"0.46530000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=77;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-23c98c88f008262185d194bd","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_max","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-24","metricId":"Accuracy","value":48.9,"rawValue":"0.489","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=69;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-a6eddd4add5a3aaff55dd0a3","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-02-19","metricId":"Accuracy","value":72.07,"rawValue":"0.7206999999999999","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=26;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-603e9e0eafa80fd0497fcbf9","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-21","metricId":"Accuracy","value":39.0,"rawValue":"0.39","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=115;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-21b0bf39edec2b397e38163a","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-05-19","metricId":"Accuracy","value":62.64000000000001,"rawValue":"0.6264000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=40;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2c998902ee64a771aa36c94f","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-21","metricId":"Accuracy","value":56.10000000000001,"rawValue":"0.561","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=53;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-795d056d0612c985da07188a","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-02-11","metricId":"Accuracy","value":48.17,"rawValue":"0.4817","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=72;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3680011d4687b16ad4d4bec7","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-07","metricId":"Accuracy","value":57.10000000000001,"rawValue":"0.5710000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=51;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3c49008818d7c650e2d88f69","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-06-16","metricId":"Accuracy","value":67.31,"rawValue":"0.6731","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=31;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-4e774ba1ed1884cfc1bc5089","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-06-16","metricId":"Accuracy","value":70.12,"rawValue":"0.7012","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=28;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f08ecad1a9135a58141123e5","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":60.70000000000001,"rawValue":"0.6070000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=44;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-5ac3c3200b4989717b6b0918","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":52.67000000000001,"rawValue":"0.5267000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=59;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-1dfd18e66bbec76879f41c66","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":38.06,"rawValue":"0.38060000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=119;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-2e7a30afbdc6fed07d0b1c90","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_low","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-08-07","metricId":"Accuracy","value":25.89,"rawValue":"0.2589","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=145;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-46c95a1cb2b5044696514e74","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-11-13","metricId":"Accuracy","value":60.77,"rawValue":"0.6077","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=43;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-7d107d2df1f1a7b2074e11dc","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_low","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy","value":49.6,"rawValue":"0.496","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=67;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-77a5a2cda57aa2c67af5e72f","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy","value":63.44,"rawValue":"0.6344","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=38;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-4a7df37769a5ffeb7a87a81c","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_none","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy","value":49.60000000000001,"rawValue":"0.49600000000000005","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=66;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-8339942025fa093de98db7e0","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-12-11","metricId":"Accuracy","value":72.19,"rawValue":"0.7219","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=25;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-33355cc4465735c14669be60","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy","value":79.3,"rawValue":"0.793","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=14;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-8be5de2765d1a0ee41911cb4","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex_xhigh","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-02-05","metricId":"Accuracy","value":77.9,"rawValue":"0.779","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=18;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-888fd7e64d89b4d91f17a03c","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_none","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-03-05","metricId":"Accuracy","value":57.440000000000005,"rawValue":"0.5744","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=49;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-a48d641af6303c46c86c6563","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-03-05","metricId":"Accuracy","value":77.7,"rawValue":"0.777","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=19;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-1516370a76499e4fc263ff8a","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-03-17","metricId":"Accuracy","value":60.3,"rawValue":"0.603","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=46;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f7cc55c6ebb2c5ccf67374c7","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_none","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-03-17","metricId":"Accuracy","value":37.9,"rawValue":"0.379","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=122;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1f890ffd1cc6393f9f11129d","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-03-17","metricId":"Accuracy","value":49.23,"rawValue":"0.49229999999999996","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=68;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0a83dca0a66c1875a63217f1","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_none","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-03-17","metricId":"Accuracy","value":37.96,"rawValue":"0.3796","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=121;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-8fee6d78fdbf8b55b06c2249","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_none","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-03-05","metricId":"Accuracy","value":57.440000000000005,"rawValue":"0.5744","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=50;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-83fb5797e29ed6eebbcc711c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy","value":83.9,"rawValue":"0.8390000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=11;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b54b7b5f9d4a33740b1d4c05","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_none","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy","value":67.15,"rawValue":"0.6715000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=33;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9c988f47aac758fafd45569e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_xhigh","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-23","metricId":"Accuracy","value":84.91,"rawValue":"0.8491","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=10;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-d169fe8b2b60050fc5689be9","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":60.86,"rawValue":"0.6086","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=42;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-96741dc153c02e43f48d03ae","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":88.76,"rawValue":"0.8876000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=6;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-65a29c0e79942adc09cbda5b","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":86.97,"rawValue":"0.8697","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=8;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-b912ba52134065df48994e7c","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_promax","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":89.43,"rawValue":"0.8943000000000001","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=5;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-858cdf9a82f6b5f1a6eb9ddf","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-09","metricId":"Accuracy","value":78.27,"rawValue":"0.7827","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=15;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2bcdec59010ff9468cb634c9","canonicalModelId":"xai/grok-4-fast@2025-09-19","modelName":"Grok 4 Fast","modelRef":"grok-4-fast","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2025-09-19","metricId":"Accuracy","value":42.86,"rawValue":"0.4286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=94;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-341ed094bfdff0baa231e202","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_unknown","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-08","metricId":"Accuracy","value":46.43,"rawValue":"0.4643","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=78;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f347c0baceb90614d1387b80","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_high","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-08-12","metricId":"Accuracy","value":67.29,"rawValue":"0.6729","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=32;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f473c1a905dcf870884661af","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-01-27","metricId":"Accuracy","value":45.6,"rawValue":"0.456","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=84;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://htihle.github.io/weirdml.html"},{"id":"pub-5bfa42cd5b40bfa478f29f04","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-20","metricId":"Accuracy","value":55.86,"rawValue":"0.5586","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=54;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-756a44a8c91a509d54363847","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-06-12","metricId":"Accuracy","value":54.120000000000005,"rawValue":"0.5412","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=57;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-020f124c2c59b561943f7f81","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-07-16","metricId":"Accuracy","value":82.57,"rawValue":"0.8257","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=13;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a9cc690a7dc306eb133c5521","canonicalModelId":"qwen/qwen3-coder-next@2026-02-03","modelName":"Qwen3-Coder-Next","modelRef":"qwen3-coder-next","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-02-02","metricId":"Accuracy","value":34.4,"rawValue":"0.344","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=139;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-be0a409296418777aa0b278f","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"qwen3.5-27B","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-02-24","metricId":"Accuracy","value":39.53,"rawValue":"0.39530000000000004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=109;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1077e16007cef6d06f4613e8","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b","benchmarkId":"epoch-weirdml_external","benchmarkVersion":"2026-04-14","metricId":"Accuracy","value":34.49,"rawValue":"0.34490000000000004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"weirdml_external.csv:row=138;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","source_version_label_without_id"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b321d60b78f3ea62eded4c12","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-10-15","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":4.137931034482759,"rawValue":"0.041379310344827586","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=65;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-16278bc1a4f0103af49e5c8e","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_32K","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-10-15","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":5.902777777777778,"rawValue":"0.059027777777777776","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=64;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f1f3e6d382f1e8987775d314","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":4.482758620689655,"rawValue":"0.04482758620689655","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=70;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f9800e442d92c647a386184a","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_27K","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":4.137931034482759,"rawValue":"0.041379310344827586","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=67;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-244fab2c69780ad856124bbb","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":20.689655172413794,"rawValue":"0.20689655172413793","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=41;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-c009d95d5bd9f4e460a8abf2","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_16K","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":20.344827586206897,"rawValue":"0.20344827586206896","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=43;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-fd0f5b1796243a99e2c6913a","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_32K","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":20.689655172413794,"rawValue":"0.20689655172413793","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=42;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3f85d87c2fb453b4ccb73b20","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":38.275862068965516,"rawValue":"0.38275862068965516","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=26;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-179e287821fc3b106aa1f568","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_32K","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":40.0,"rawValue":"0.4","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=25;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-e3218a0b0944ad986a30e2c1","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_64K","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":39.6551724137931,"rawValue":"0.39655172413793105","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=24;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b76a51b69fb2f42d973d56cc","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":40.699999999999996,"rawValue":"0.407","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=23;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-3d26ba3120baa77358baf6b9","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_xhigh","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":43.79310344827586,"rawValue":"0.4379310344827586","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=14;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8de509232aee018c87ee66cb","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-05-28","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":47.241379310344826,"rawValue":"0.4724137931034483","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=2;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c5e057d430672fc0d39ca03f","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":9.310344827586208,"rawValue":"0.09310344827586207","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=50;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-aa1e8961b2db1d9450869636","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":15.22491349480969,"rawValue":"0.1522491349480969","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=51;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-3651a98edba24e0434e649e3","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_59K","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":13.494809688581316,"rawValue":"0.13494809688581316","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=56;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-71b34ec3865f180c97542368","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_16K","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-02-17","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":32.4,"rawValue":"0.324","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=20;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bff9ae4f39c576cf377ba322","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"fireworks/deepseek-v3p2","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-12-01","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":22.1,"rawValue":"0.221","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=29;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-19d061b04065c0aa73992268","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"fireworks/kimi-k2p5","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-01-27","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":27.900000000000002,"rawValue":"0.279","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=27;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-50ef7ef9e24438145e5d0824","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"gemini-2.5-flash","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-06-17","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":4.844290657439446,"rawValue":"0.04844290657439446","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=30;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b009fdc29c1e261afbc4bd51","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-06-17","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":14.13793103448276,"rawValue":"0.1413793103448276","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=44;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2a2a927857bef3d0f29bc87e","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-02-19","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":36.9,"rawValue":"0.369","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=22;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f76ab8b45ad0fcbd4bfa9b3a","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":38.96551724137931,"rawValue":"0.3896551724137931","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=4;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6972409a2ce270b48e96b0c1","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-02-11","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":16.433566433566433,"rawValue":"0.16433566433566432","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=21;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-172dd16f4db4a31bee79a406","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-04-07","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":33.44827586206897,"rawValue":"0.33448275862068966","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=9;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-691c64ddb49bef3dc9b49fc8","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":32.41379310344827,"rawValue":"0.32413793103448274","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=55;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-b50c5abbcbd2c7de889c58eb","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":27.241379310344826,"rawValue":"0.27241379310344827","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=62;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1c2caf9cad755c097950f98c","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":27.241379310344826,"rawValue":"0.27241379310344827","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=57;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c66063457db5bab8b413d306","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":20.344827586206897,"rawValue":"0.20344827586206896","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=60;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ba7029ce10b2ab5338bb1bd4","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":8.275862068965518,"rawValue":"0.08275862068965517","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=63;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-c6ec7c2838844c157df7c2d5","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":7.241379310344828,"rawValue":"0.07241379310344828","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=66;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9c536a67d10b060556a6c65f","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":31.03448275862069,"rawValue":"0.3103448275862069","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=54;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b85ec19119b0ce9fa34de18a","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_low","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":17.301038062283737,"rawValue":"0.17301038062283736","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=40;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-790020830060d671f0a82857","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":26.89655172413793,"rawValue":"0.2689655172413793","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=46;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-0b0bf2860cad8a0256fdd69e","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_none","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":2.097902097902098,"rawValue":"0.02097902097902098","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=39;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-071819aeeed98d8c7d2821bd","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":40.3448275862069,"rawValue":"0.40344827586206894","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=33;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-a8c9cc84832d82599340a295","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_low","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":26.551724137931032,"rawValue":"0.2655172413793103","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=35;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-d1152a30a69a82f52ad85a68","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":36.896551724137936,"rawValue":"0.3689655172413793","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=34;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bda1f9093a27d85ed88f4fce","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":40.699999999999996,"rawValue":"0.407","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=32;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-5f9112587dbb0c00315cad64","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":47.599999999999994,"rawValue":"0.476","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=18;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-18d300317eb4909b9b3f6593","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":28.28,"rawValue":"0.2828","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=15;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c0297a540201a8f071c6488f","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":25.86,"rawValue":"0.2586","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=16;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-14612a3c3f64ab64aa677261","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_xhigh","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":50.0,"rawValue":"0.5","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=19;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-33827105735abbf721133218","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-04-20","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":38.96551724137931,"rawValue":"0.3896551724137931","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=10;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-cae9720fdc1e866d14a0165a","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-02-25","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":6.206896551724138,"rawValue":"0.06206896551724138","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=8;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2c5096d3e72a21c1a3349a16","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-02-16","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":21.03448275862069,"rawValue":"0.2103448275862069","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=5;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b46a82075d9ce4c0b83c2613","canonicalModelId":"qwen/qwen3.6-flash@2026-04-27","modelName":"Qwen3.6 Flash","modelRef":"qwen3.6-flash","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-04-27","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":10.344827586206897,"rawValue":"0.10344827586206896","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=7;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-328c33c5e10bcbbbb29a46d7","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-04-20","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":23.103448275862068,"rawValue":"0.23103448275862068","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=3;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-98d867e21180d8bc37fd3248","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2026-03-31","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":26.20689655172414,"rawValue":"0.2620689655172414","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=6;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-deb3761ae298f4785b0bbd76","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/GLM-4.7","benchmarkId":"frontiermath","benchmarkName":"FrontierMath T1–T3","benchmarkVersion":"2025-12-22","benchmarkVersionHint":"frontiermath@v2-t1-t3","metricId":"mean_score","value":2.4390243902439024,"rawValue":"0.024390243902439025","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=28;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4fe3042119355255cc9c11e5","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-05-28","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":76.32575757575758,"rawValue":"0.7632575757575758","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=160;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-921524d4f34b011ec22e943e","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMax-M3","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-06-01","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":90.9090909090909,"rawValue":"0.9090909090909091","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=8;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8b6b9e7f732e37963801eb54","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"MiniMaxAI/MiniMax-M2.1","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":80.8080808081,"rawValue":80.8080808081,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=63;row=62","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa-diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/08/7b/087b0dc6-3c87-4f40-8458-16f19de92200.json"},{"id":"pub-a7e89b1f4e9a5fbe4fa482a2","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMaxAI/MiniMax-M2.5","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":85.2,"rawValue":85.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=41;row=40","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M2.5"},{"id":"pub-549c41eb2cf8950f7e9eb3eb","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"Qwen/Qwen3.5-122B-A10B","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":86.6,"rawValue":86.6,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=29;row=28","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa_1.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-122B-A10B"},{"id":"pub-01452c192b3b2c2863a44824","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"Qwen/Qwen3.5-122B-A10B","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":84.8484848485,"rawValue":84.8484848485,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=43;row=42","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa-diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/58/60/5860f5bd-926e-4ece-bc87-a1af551a6280.json"},{"id":"pub-ebf26d21d1bac77bf23ba7e7","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"Qwen/Qwen3.5-27B","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":85.5,"rawValue":85.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=39;row=38","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa_1.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-27B"},{"id":"pub-1787df51008565c9f6882080","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"Qwen/Qwen3.5-27B","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":81.8181818182,"rawValue":81.8181818182,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=59;row=58","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa-diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/5b/a4/5ba4cf19-18b3-4d08-bad0-ef29422cb18a.json"},{"id":"pub-9b5b1e46f0a27386bda9717a","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"Qwen/Qwen3.5-397B-A17B","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":88.4,"rawValue":88.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=15;row=14","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa_diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-397B-A17B"},{"id":"pub-ef8f15427e7456b63f39ff6b","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"Qwen/Qwen3.6-27B","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":87.8,"rawValue":87.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=22;row=21","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-27B"},{"id":"pub-4e3425007ca54f34dddb8ac3","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"Qwen/Qwen3.6-35B-A3B","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":86.0,"rawValue":86,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=35;row=34","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-35B-A3B"},{"id":"pub-2cb08742678a21b103739b81","canonicalModelId":"qwen/qwen3.8-2.4t-a95b@2026-08-12","modelName":"Qwen3.8 2.4T A95B","modelRef":"Qwen/Qwen3.8-2.4T-A95B","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":92.6,"rawValue":92.6,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=3;row=2","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B"},{"id":"pub-6919e4db816f8d64a713bc5a","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"Qwen/Qwen3.8-27B","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":89.2,"rawValue":89.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=13;row=12","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/Qwen3.8-27B.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.8-27B"},{"id":"pub-5933de1179e8edb153efdfc6","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"Qwen/Qwen3.8-Flash-Next","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":91.7,"rawValue":91.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=4;row=3","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/Qwen3.8-Flash-Next.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.8-Flash-Next"},{"id":"pub-872a1f364f5dece8e63f2e63","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"XiaomiMiMo/MiMo-V2-Flash","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":83.3333333333,"rawValue":83.3333333333,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=52;row=51","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa-diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/36/22/3622b785-41c0-4a96-b867-5bde46851587.json"},{"id":"pub-669d356be453b229ccde054d","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"XiaomiMiMo/MiMo-V2.5-Pro","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":66.7,"rawValue":66.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=93;row=92","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro"},{"id":"pub-ef5d1f3a7ccef5947cf20e89","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-06-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":83.33333333333334,"rawValue":"0.8333333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=70;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-eeeb81aa467b0f4ca561cd52","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-06-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":78.78787878787878,"rawValue":"0.7878787878787878","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=69;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-e8c100081b9dc12e6eacc395","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-06-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":85.85858585858585,"rawValue":"0.8585858585858586","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=68;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-ce414f86948ba83d3b9803d8","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-10-15","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":60.47979797979798,"rawValue":"0.6047979797979798","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=148;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-64b456bd45990904d2c68569","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_32K","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-10-15","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":71.21212121212122,"rawValue":"0.7121212121212122","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=146;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d3815943035ba8563ee37a12","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":69.1919191919192,"rawValue":"0.6919191919191919","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=170;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4ff5876157472d164455c430","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_16K","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":76.26262626262627,"rawValue":"0.7626262626262627","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=167;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-3a0b516b6a34008903657994","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":80.68181818181817,"rawValue":"0.8068181818181818","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=134;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-c71134eb80c7ee0f3bd513a7","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_16K","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":85.47979797979798,"rawValue":"0.8547979797979798","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=132;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d30f283391a157755bf30ae1","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_32K","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":86.0479797979798,"rawValue":"0.860479797979798","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=133;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5f08c08d27f0b5b80e3252fc","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_32K","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":90.53030303030303,"rawValue":"0.9053030303030303","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=120;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-30a8c1a35c9d3e5129cc91f7","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_64K","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":88.76262626262627,"rawValue":"0.8876262626262627","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=121;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-c6ba0cf59dd68267c2e66dc6","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":88.38383838383838,"rawValue":"0.8838383838383839","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=49;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f762b8966beca6fd93e61e1c","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":86.36363636363636,"rawValue":"0.8636363636363636","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=67;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1b444cb3f72cdcc8594d0c6b","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_xhigh","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":90.15151515151516,"rawValue":"0.9015151515151515","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=111;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-dad525f30663a32c8629db86","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-05-28","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":88.38383838383838,"rawValue":"0.8838383838383839","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=66;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-e569922d07f2891afa460452","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-05-28","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":91.03535353535354,"rawValue":"0.9103535353535354","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=105;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-010d54fff00ef7226d59b01e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-05-28","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":85.35353535353535,"rawValue":"0.8535353535353535","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=65;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-de383f1bd8b384fff42a5131","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-24","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":92.92929292929293,"rawValue":"0.9292929292929293","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=63;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-42fa8ec29f4bb7e2eef887fe","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-24","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":87.87878787878788,"rawValue":"0.8787878787878788","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=64;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-293dc4c10503df5943b1fb9c","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-24","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":93.87626262626263,"rawValue":"0.9387626262626263","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=78;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-504da72f32cb3f60cc39bbae","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":73.73737373737373,"rawValue":"0.7373737373737373","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=150;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-a2a7ac9b402de73084096006","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_16K","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":78.78787878787878,"rawValue":"0.7878787878787878","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=143;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-bc1c455b80d920790e7b106c","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_32K","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":81.7258883248731,"rawValue":"0.817258883248731","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=147;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-418291204b36d7c753c7f083","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_59K","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":82.32323232323232,"rawValue":"0.8232323232323232","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=144;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-71211b0b70d5a8fbcc93de61","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_32K","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-17","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":87.37373737373737,"rawValue":"0.8737373737373737","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=117;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-53f2ce9a365f458875a0c5ad","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-17","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":83.33333333333334,"rawValue":"0.8333333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=91;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-85dd4ef43c66de58a71a5339","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-17","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":78.78787878787878,"rawValue":"0.7878787878787878","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=62;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-80ac6b5c90318ee875adfb3d","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_medium","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-17","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":83.33333333333334,"rawValue":"0.8333333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=94;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-48ff75929bdc171e3f97eae5","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-06-30","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":80.3030303030303,"rawValue":"0.803030303030303","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=61;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-43e0ac11a075aad69dff57f8","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_xhigh","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-06-30","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":90.53030303030303,"rawValue":"0.9053030303030303","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=101;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-f4b5ce6a8ab5d45cd89a9c64","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"deepseek-ai/DeepSeek-R1-0528","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":78.7878787879,"rawValue":78.7878787879,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=74;row=73","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa_diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/a1/dd/a1dd2ea2-f092-423e-bc53-04c2fba5948d.json"},{"id":"pub-7f770c6a659f6e05ac3f7519","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-ai/DeepSeek-V4-Flash","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":88.1,"rawValue":88.1,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=19;row=18","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash"},{"id":"pub-213e986a2f07ced2aef02423","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-ai/DeepSeek-V4-Pro","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":90.1,"rawValue":90.1,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=10;row=9","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro"},{"id":"pub-2fb198af14e00fd4f3cab7a5","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"deepseek-chat","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-12-01","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":71.21212121212122,"rawValue":"0.7121212121212122","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=83;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-01","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0ebdfb9a0bed05d3f3e1a8ea","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-31","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":91.03535353535354,"rawValue":"0.9103535353535354","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=76;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c41eebd48ab17f806c3d3261","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-08-13","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":91.66666666666666,"rawValue":"0.9166666666666666","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=3;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-93832bafdc60ffb6d2648eef","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-24","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":90.9090909090909,"rawValue":"0.9090909090909091","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=48;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-7fda6b6774934380a3289bba","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-24","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":89.64646464646465,"rawValue":"0.8964646464646465","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=104;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-a3cc6e82dde967056ac7c83d","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-24","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":73.23232323232324,"rawValue":"0.7323232323232324","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=47;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0cbac08478a0835c4a67d714","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"fireworks/kimi-k2p5","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-01-27","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":87.6,"rawValue":"0.876","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=122;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c3971f1e4fdaf1c9aaee4b9f","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-06-17","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":85.29040404040404,"rawValue":"0.8529040404040404","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=137;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-aebbfcb47ec31709acb564a3","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-19","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":94.1,"rawValue":"0.941","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=118;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-50fac8d3b1963045dc8a786a","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-19","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":94.44444444444444,"rawValue":"0.9444444444444444","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=56;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ec5a110adae2297425fa20b2","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":83.33333333333334,"rawValue":"0.8333333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=54;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d72cb63d93dcca012e888fbd","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":75.75757575757575,"rawValue":"0.7575757575757576","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=52;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-852f97c42d88f1c0044f3365","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite_minimal","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":74.24242424242425,"rawValue":"0.7424242424242424","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=53;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-33c9238f3ba1df61f8fb94cc","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":92.8030303030303,"rawValue":"0.928030303030303","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=106;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-03e83b6c267809a4944578c2","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":88.88888888888889,"rawValue":"0.8888888888888888","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=55;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-1863a498d5c035e2ae1d8c37","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_minimal","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":86.36363636363636,"rawValue":"0.8636363636363636","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=89;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-097b31e0df114d9fce5824ae","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":94.12878787878788,"rawValue":"0.9412878787878788","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=77;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-42d2e1823ee321552b676b6b","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":86.36363636363636,"rawValue":"0.8636363636363636","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=51;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-8949cb7681a3e8ac0db9408d","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_minimal","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":85.85858585858585,"rawValue":"0.8585858585858586","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=33;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c0154680177237a53b09e0ed","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-08-13","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":94.82323232323232,"rawValue":"0.9482323232323232","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=4;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1b10511e22040e7e61e24902","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-12-22","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":83.33333333333334,"rawValue":"0.8333333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=123;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-77448370ef290092e70f92c9","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-11","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":87.81725888324873,"rawValue":"0.8781725888324873","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=119;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-98f3bbbfeb7753d4e81a9b95","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-07","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":89.8989898989899,"rawValue":"0.898989898989899","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=10;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-183ae87c492ebef6e3e03ec8","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-06-16","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":91.85606060606061,"rawValue":"0.9185606060606061","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=102;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-64c1e22f2613a0aa672ddb1f","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-06-16","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":71.21212121212122,"rawValue":"0.7121212121212122","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=9;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-557a3d96294f8737cef7d950","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-08-14","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":90.9090909090909,"rawValue":"0.9090909090909091","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=2;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c84bceb2fe13ad97e717268a","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":86.17424242424242,"rawValue":"0.8617424242424242","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=142;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2f13ffe0f9cd433b1f6ba7ac","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":85.35353535353535,"rawValue":"0.8535353535353535","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=152;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-976dbb39ac14a38a5127e285","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_minimal","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":71.71717171717171,"rawValue":"0.7171717171717171","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=80;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-287a00fdca1dcd34985457c8","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":75.0,"rawValue":"0.75","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=141;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-9479d715c79d62d23782ce0e","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":71.65404040404042,"rawValue":"0.7165404040404041","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=151;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d8375b54cf47da9e2e76abdd","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_minimal","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":71.71717171717171,"rawValue":"0.7171717171717171","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=44;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-84174717ccb9e5ff7e8d40e6","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":69.44444444444444,"rawValue":"0.6944444444444444","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=140;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-077797ea67384b710b20ed94","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":57.57575757575758,"rawValue":"0.5757575757575758","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=90;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-defa91fe0ae123728827d062","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_medium","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":67.42424242424242,"rawValue":"0.6742424242424242","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=153;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-63f4b9409aa0159db6bfd36c","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_minimal","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":48.484848484848484,"rawValue":"0.48484848484848486","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=92;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-babf1cd0fc9af0e0b865c09a","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":87.62626262626263,"rawValue":"0.8762626262626263","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=138;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3d26d423bf0cae184ffeb62a","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_medium","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":85.03787878787878,"rawValue":"0.8503787878787878","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=136;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-2fa0d1ff70ffcbeb3169d48f","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":66.66666666666666,"rawValue":"0.6666666666666666","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=22;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-57933c43e351425c4d88286e","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":88.19444444444444,"rawValue":"0.8819444444444444","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=129;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-828e9e8136a517a6f58d1a46","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":82.7020202020202,"rawValue":"0.827020202020202","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=131;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-d6a007eb77c47a28bb619eb3","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":87.87878787878788,"rawValue":"0.8787878787878788","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=130;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0183d18bb4f56e5a2be2fe5a","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":73.23232323232324,"rawValue":"0.7323232323232324","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=93;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-9a9b027cf977971e1ef6a603","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":91.4,"rawValue":"0.914","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=126;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-149c46f77c9ec1e37e384bb7","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":89.8989898989899,"rawValue":"0.898989898989899","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=87;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":5},{"id":"pub-5b4fddf3504c0bd2d0e74c00","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":84.84848484848484,"rawValue":"0.8484848484848485","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=88;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-d0f662a790f2182e67f0158d","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_medium","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":88.88888888888889,"rawValue":"0.8888888888888888","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=85;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-531f4d0b0ba88a9badcb7357","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":74.74747474747475,"rawValue":"0.7474747474747475","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=86;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-4dce1779968c1d8eb6a330e2","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":93.30000000000001,"rawValue":"0.933","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=116;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-216d08b9c58b2877cc2e3487","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":83.58585858585859,"rawValue":"0.8358585858585859","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=112;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-4059d21ee5fa18b0b606eea2","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":64.14141414141415,"rawValue":"0.6414141414141414","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=17;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ec259e332df90d4223a08fe8","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_xhigh","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":86.86868686868688,"rawValue":"0.8686868686868687","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=11;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a1dc6f35831aa78b913e9e77","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":78.47,"rawValue":"0.7847","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=113;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2944506b62e65f736039c126","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":72.22222222222221,"rawValue":"0.7222222222222222","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=45;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-9e5815eca216d7268485f77c","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":55.55555555555556,"rawValue":"0.5555555555555556","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=42;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b9d1ea64ecf7180e8e63deaa","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_xhigh","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":94.6,"rawValue":"0.946","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=115;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dab8c8a35a0b21a34b015573","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-05-05","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":82.51262626262627,"rawValue":"0.8251262626262627","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=75;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-736d5071cfee0c8bb863e192","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-23","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":90.65656565656566,"rawValue":"0.9065656565656566","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=107;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-f5432d156e3b9e881ade1a1c","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-23","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":77.27272727272727,"rawValue":"0.7727272727272727","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=13;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-23","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1f8b2e40563b6fa456d36cad","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":82.32323232323232,"rawValue":"0.8232323232323232","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=37;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-a73a3be11b5eaa90eb8baf58","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":91.60353535353535,"rawValue":"0.9160353535353535","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=98;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-5a68a92215ded13d15a85bae","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":63.63636363636363,"rawValue":"0.6363636363636364","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=24;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-2ad58c47659439a3d8b690ca","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":89.8989898989899,"rawValue":"0.898989898989899","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=19;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-4884438b02da1cb2d7c1e622","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":93.49747474747475,"rawValue":"0.9349747474747475","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=99;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-440bb0b598aab9d93e79fc60","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":82.82828282828282,"rawValue":"0.8282828282828283","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=12;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b45ca376081c3ec78c83d801","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":87.37373737373737,"rawValue":"0.8737373737373737","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=25;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2e28c81b2de675137e7dbb6c","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":93.3080808080808,"rawValue":"0.9330808080808081","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=97;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-4c43956fc0c1e162c878832e","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":77.27272727272727,"rawValue":"0.7727272727272727","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=16;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-2d4cfc078d2bab08f594c204","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-08","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":93.43434343434343,"rawValue":"0.9343434343434344","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=100;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a1a617a203f4fb8c4c69b441","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-08-12","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":94.00252525252525,"rawValue":"0.9400252525252525","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=7;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9aed01fcdc809785488433af","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_xhigh","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-08-12","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":93.18181818181817,"rawValue":"0.9318181818181818","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=6;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-04bfb0d2efe2cc62c595c9ac","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-20","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":90.78282828282829,"rawValue":"0.9078282828282829","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=108;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-38b30103af59a4dba627d845","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-06-12","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":87.87878787878788,"rawValue":"0.8787878787878788","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=23;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c5e9cfba01527c651e0fc803","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_high","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-16","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":91.91919191919192,"rawValue":"0.9191919191919192","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=15;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-39f78d807b6ed31131398a36","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_low","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-16","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":84.84848484848484,"rawValue":"0.8484848484848485","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=14;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-1be29958869492fd64aff4d6","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-16","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":93.11868686868688,"rawValue":"0.9311868686868687","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=82;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":3},{"id":"pub-a5a70bfb2cd6e11e556b73a7","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"moonshotai/Kimi-K2.5","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":87.6,"rawValue":87.6,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=23;row=22","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa_diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.5"},{"id":"pub-6c7296d89a6c8c23da91d62c","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"moonshotai/Kimi-K2.5","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":87.3737373737,"rawValue":87.3737373737,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=24;row=23","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa-diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/5c/4b/5c4b9bd9-14db-4c11-bf99-9661461459f6.json"},{"id":"pub-3234f5c09497f3829d541590","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"moonshotai/Kimi-K2.6","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":90.5,"rawValue":90.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=7;row=6","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa_diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6"},{"id":"pub-5a804e9c0fbc819811228413","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"moonshotai/Kimi-K3","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":93.5,"rawValue":93.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=1;row=0","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K3"},{"id":"pub-82be1d46353642a44ebaa477","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-13","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":85.85858585858585,"rawValue":"0.8585858585858586","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=30;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-a678294d6f8bce35e76eb0a5","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"qwen3.5-397b-a17b_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-13","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":86.36363636363636,"rawValue":"0.8636363636363636","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=31;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d4506121da75b0ed87b8914c","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-25","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":82.32323232323232,"rawValue":"0.8232323232323232","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=43;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9e9fd41e03e75a4c853d11e9","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-02-16","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":84.84848484848484,"rawValue":"0.8484848484848485","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=32;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e1e6bfc3d2fe92af603ba8ef","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-22","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":85.85858585858585,"rawValue":"0.8585858585858586","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=28;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5a32af5f0fbcabca81f4b43d","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-22","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":84.84848484848484,"rawValue":"0.8484848484848485","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=29;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-9bfacb06f72b0d116612b3ad","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-14","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":83.83838383838383,"rawValue":"0.8383838383838383","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=34;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-25735c1c2f2e026a589bfea2","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"qwen3.6-35b-a3b_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-14","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":84.84848484848484,"rawValue":"0.8484848484848485","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=35;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-bce7b2ebd024eeff4ac10f32","canonicalModelId":"qwen/qwen3.6-flash@2026-04-27","modelName":"Qwen3.6 Flash","modelRef":"qwen3.6-flash","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-27","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":83.33333333333334,"rawValue":"0.8333333333333334","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=36;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-55720cbace065dfe3264e9bf","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-04-20","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":87.37373737373737,"rawValue":"0.8737373737373737","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=18;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6e3812d40a0e6b3886bc49f6","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-03-31","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":88.38383838383838,"rawValue":"0.8838383838383839","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=26;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-144d4c7cfc6339020cd6e3e6","canonicalModelId":"qwen/qwen3.7-flash@2026-07-27","modelName":"Qwen3.7 Flash","modelRef":"qwen3.7-flash","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-27","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":82.32323232323232,"rawValue":"0.8232323232323232","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=40;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-6865bbb19e52f350dd9400ca","canonicalModelId":"qwen/qwen3.7-flash@2026-07-27","modelName":"Qwen3.7 Flash","modelRef":"qwen3.7-flash_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-07-27","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":80.8080808080808,"rawValue":"0.8080808080808081","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=41;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8cb8550da98d0bc85a39c737","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":90.9090909090909,"rawValue":"0.9090909090909091","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=20;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-573447d84bd59bd366bc182a","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-06-02","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":87.87878787878788,"rawValue":"0.8787878787878788","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=38;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-08cb6e1981375ff57beb0dc0","canonicalModelId":"qwen/qwen3.7-plus@current","modelName":"Qwen3.7 Plus","modelRef":"qwen3.7-plus_none","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-06-02","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":81.81818181818183,"rawValue":"0.8181818181818182","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=39;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-3635258e703dd30f599376c8","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max_xhigh","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersion":"2026-08-02","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"mean_score","value":92.67676767676768,"rawValue":"0.9267676767676768","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=74;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bf0bc9eb72aecff640badabd","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/GLM-4.7","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":85.7,"rawValue":85.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=37;row=36","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/zai-org/GLM-4.7"},{"id":"pub-2e2a183c67063a382fbb1aa0","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/GLM-4.7-FP8","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":87.8787878788,"rawValue":87.8787878788,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=21;row=20","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa-diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/97/fa/97faabc6-a7ca-4cad-8d29-307103287d91.json"},{"id":"pub-ec8d4fb1af24105a3595d65a","canonicalModelId":"zai/glm-4.7-flash@current","modelName":"GLM-4.7 Flash","modelRef":"zai-org/GLM-4.7-Flash","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":75.2,"rawValue":75.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=79;row=78","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-4.7-Flash"},{"id":"pub-a89621805e42c57d6e012792","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"zai-org/GLM-5","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":86.0,"rawValue":86,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=34;row=33","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5"},{"id":"pub-c1d631e567927882fffd486f","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"zai-org/GLM-5-FP8","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":85.3535353535,"rawValue":85.3535353535,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=40;row=39","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa-diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/8a/87/8a87d622-4abb-48a7-8ac3-594e31a85525.json"},{"id":"pub-02f6b891343a5142707be676","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"zai-org/GLM-5.1","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":86.2,"rawValue":86.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=33;row=32","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.1"},{"id":"pub-a6412a3a306c5d5292a7c9b8","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"zai-org/GLM-5.1-FP8","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":83.3333333333,"rawValue":83.3333333333,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=53;row=52","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa-diamond.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/52/23/522348c0-d242-4517-b97c-4e55c4261882.json"},{"id":"pub-2bff2225f45a5bdbff456a24","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"zai-org/GLM-5.2","benchmarkId":"gpqa-diamond","benchmarkName":"GPQA Diamond","benchmarkVersionHint":"gpqa-diamond@v1","metricId":"score","value":91.2,"rawValue":91.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=5;row=4","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","protocol":{"filename":".eval_results/gpqa.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.2"},{"id":"pub-e2083acf3133e640402a403f","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.6659192825112108,"rawValue":0.6659192825112108,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=36;column=GPQA - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7428cfc243518f3b9e1953a5","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"eval","value":446.0,"rawValue":446.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=GPQA - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-f100a3b9bdf0a3a58aff8ef6","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":16.325411326249803,"rawValue":16.325411326249803,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=36;column=GPQA - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-108015c827a8bf76f6f03075","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":453.1434977578475,"rawValue":453.1434977578475,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=GPQA - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c89a94617919db33cbe30c20","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":250.73766816143498,"rawValue":250.73766816143498,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=GPQA - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-e057da1c0199c605b046e619","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=GPQA - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-28a4fb57ea854c1b289cdd79","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=GPQA - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-49285b3e172b82c65ce65816","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.7085201793721974,"rawValue":0.7085201793721974,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=37;column=GPQA - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-81d428f5396c2750b51e95cb","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"eval","value":446.0,"rawValue":446.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=GPQA - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-55f58771a7f4ac232527b17c","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":45.52923426562793,"rawValue":45.52923426562793,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=37;column=GPQA - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-233ba6da46d0adc76c4bdede","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":343.76233183856505,"rawValue":343.76233183856505,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=GPQA - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-ad03caa7ed7616251b512d01","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":272.73766816143495,"rawValue":272.73766816143495,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=GPQA - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1ade467dde068dcb75fee7de","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=GPQA - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-85a959feb68ca147fc5afcc8","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=GPQA - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3375c76b2f233f4d95129788","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.6053811659192825,"rawValue":0.6053811659192825,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=39;column=GPQA - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e5031dba02ec96a38620a072","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"eval","value":446.0,"rawValue":446.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=GPQA - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-b196ebf0d74ac8896fd9a076","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":5.102193982611857,"rawValue":5.102193982611857,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=39;column=GPQA - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-b3a5052474884f3abdd6b536","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":524.5246636771301,"rawValue":524.5246636771301,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=GPQA - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-6c2cc5b09ad56e48873bd81e","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":272.73766816143495,"rawValue":272.73766816143495,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=GPQA - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c271f8a8443a8cd54eb24173","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=GPQA - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-9a427d9e76c479e4b9b3d0c7","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=GPQA - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-49aef1f922617d345fee6bd3","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.6860986547085202,"rawValue":0.6860986547085202,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=38;column=GPQA - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f329394ab55db45089725c39","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"eval","value":446.0,"rawValue":446.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=GPQA - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-b92e7ba07f20f0c08f5e7518","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":12.414452127318263,"rawValue":12.414452127318263,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=38;column=GPQA - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-82c8b45c20c6b3fc4c7dec53","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":544.2152466367713,"rawValue":544.2152466367713,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=GPQA - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-65526b84cffa4596621143b3","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":272.73766816143495,"rawValue":272.73766816143495,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=GPQA - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-7800a2b48b840cfb073582ec","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=GPQA - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c47347eae53e0e1c8d613608","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=GPQA - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-08d12d4a00361224c17d3ba0","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.6659192825112108,"rawValue":0.6659192825112108,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=25;column=GPQA - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-46dbf1518771d86f75e5e2b2","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"eval","value":446.0,"rawValue":446.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=GPQA - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-d161439fe4613a69e065e194","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":155.438512681311,"rawValue":155.438512681311,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=25;column=GPQA - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-fe76ba2bf1f4445be4ea1933","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=GPQA - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-d0961d4e356385c6bcaa7b70","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":261.5896860986547,"rawValue":261.5896860986547,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=GPQA - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-ee5451ace7d34b29ff5fae5b","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=GPQA - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3f951470206a8dfb86aa08de","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=GPQA - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-96eee1d6ecd0b027dff05aa9","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.7914798206278026,"rawValue":0.7914798206278026,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=55;column=GPQA - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-5827fc592b7009fdac4c9a34","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"eval","value":446.0,"rawValue":446.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=GPQA - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-16f13900caae2ac83bf7f137","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":57.41822674028542,"rawValue":57.41822674028542,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=55;column=GPQA - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-0938a8b2440031aaf2a674ae","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":5.934977578475336,"rawValue":5.934977578475336,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=GPQA - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1e47ee2f81b36024014184a2","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":268.15246636771303,"rawValue":268.15246636771303,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=GPQA - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-756a5f3bdf51e020fd07b4b9","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=GPQA - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-7daee7e7e08e33b813c1fd06","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=GPQA - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-91b18b348ee385e0fe48fb97","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.7556053811659192,"rawValue":0.7556053811659192,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=56;column=GPQA - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-87617270986c06eac6c9853b","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"eval","value":446.0,"rawValue":446.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=GPQA - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3441a3d70e17fa0d5499ae0a","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":20.737325443280653,"rawValue":20.737325443280653,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=56;column=GPQA - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-b168dba2ceb4ab0a86c2897d","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":25.378923766816143,"rawValue":25.378923766816143,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=GPQA - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-05deb1a96744157a1417652a","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":268.15246636771303,"rawValue":268.15246636771303,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=GPQA - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-92fdad27cfd634dbb88437ea","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=GPQA - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-591fb7af0bac075673eca625","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=GPQA - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c220f6ba264611824b198ed1","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.679372197309417,"rawValue":0.679372197309417,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=57;column=GPQA - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4ab75c16ca46703180621a53","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"eval","value":446.0,"rawValue":446.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=GPQA - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-bd82dbbdf0b81457f210415d","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":30.2457077674267,"rawValue":30.2457077674267,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=57;column=GPQA - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-d24f14f769d68fd4caa02e37","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":5.668161434977579,"rawValue":5.668161434977579,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=GPQA - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-4562007455df79ef681da2d7","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":268.15246636771303,"rawValue":268.15246636771303,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=GPQA - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-084d6e6692f02a5ddd2a6996","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=GPQA - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-415deae519014f5cad8e569a","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=GPQA - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1c45812cf388e91df9bb0915","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.44170403587443946,"rawValue":0.44170403587443946,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=58;column=GPQA - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-67140daf2514422e91be6b46","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"eval","value":446.0,"rawValue":446.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=GPQA - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-ba73b9268d8dd6cfd63c24b1","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":1.002433323539426,"rawValue":1.002433323539426,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=58;column=GPQA - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-06d64dd174bc80d7eac00d89","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":5.42152466367713,"rawValue":5.42152466367713,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=GPQA - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3a5af8cde5fd6b8663a4df5d","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":268.15246636771303,"rawValue":268.15246636771303,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=GPQA - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-18b02e2b1221c1f52d3d7a37","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=GPQA - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-52b97e0e0a358c4b9196db88","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-gpqa","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=GPQA - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-9aeb51448163037c677b68f2","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"eval","value":541.0,"rawValue":541.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=IFEval - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-7d681990442fe586a7076bee","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"ifeval-strict-acc","value":0.9177449168207027,"rawValue":0.9177449168207027,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=36;column=IFEval - IFEval Strict Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f02e6da48f010d58e989600e","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"observed-inference-time-s","value":16.576411149939712,"rawValue":16.576411149939712,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=36;column=IFEval - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-7e996bca1b26ac5a68cee407","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"output-tokens","value":422.7744916820702,"rawValue":422.7744916820702,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=IFEval - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-12d61ad4da00cf8165b06a67","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"prompt-tokens","value":47.15896487985213,"rawValue":47.15896487985213,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=IFEval - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c3697aa048300c6d8441529e","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=IFEval - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-285239fcd6c3d9ca827fea88","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=IFEval - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c8061ce3b588b992be328236","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"eval","value":541.0,"rawValue":541.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=IFEval - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-9bbbc8350ddae34f39722927","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"ifeval-strict-acc","value":0.8487369069624158,"rawValue":0.8487369069624158,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=37;column=IFEval - IFEval Strict Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dd24ca76800ad1cbb67f6774","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"observed-inference-time-s","value":22.45251508421368,"rawValue":22.45251508421368,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=37;column=IFEval - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a35d66110c82915fbf180f97","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"output-tokens","value":403.74491682070243,"rawValue":403.74491682070243,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=IFEval - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-18d7a44d4b8e457e22ac9db2","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"prompt-tokens","value":47.15896487985213,"rawValue":47.15896487985213,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=IFEval - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-0fc4c7a16e767edff5bc6bf4","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=IFEval - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a41d1b8120b65c93e80c0cb3","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=IFEval - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-4c568f7cd834f9a264bcaf6e","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"eval","value":541.0,"rawValue":541.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=IFEval - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a8f6e4d968ddcdf077e12690","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"ifeval-strict-acc","value":0.8012939001848433,"rawValue":0.8012939001848433,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=39;column=IFEval - IFEval Strict Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ea792de09db76906375e49b0","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"observed-inference-time-s","value":4.355410516372229,"rawValue":4.355410516372229,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=39;column=IFEval - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-68c85942a8c524e73833027d","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"output-tokens","value":390.4158964879852,"rawValue":390.4158964879852,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=IFEval - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a042d48bc744986c15b6456d","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"prompt-tokens","value":47.15896487985213,"rawValue":47.15896487985213,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=IFEval - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-81f45c906f303ef4f3080bb5","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=IFEval - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-ff44ae6aeea29ab99e9583c0","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=IFEval - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1748a54b54d7799e25b983e2","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"eval","value":541.0,"rawValue":541.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=IFEval - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1aba7518f77cc2f9648e43cc","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"ifeval-strict-acc","value":0.8496611213801607,"rawValue":0.8496611213801607,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=38;column=IFEval - IFEval Strict Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2931b17f65d2edcffc265e2e","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"observed-inference-time-s","value":10.90394415211986,"rawValue":10.90394415211986,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=38;column=IFEval - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-99e7b7294ca7f8eff060861a","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"output-tokens","value":414.63216266173754,"rawValue":414.63216266173754,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=IFEval - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-4e21fa39846726355a7d2fd6","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"prompt-tokens","value":47.15896487985213,"rawValue":47.15896487985213,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=IFEval - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-df619799122a20091ac2e610","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=IFEval - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-11074874bb4ad65c8031aae3","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=IFEval - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-f9cc90e25dd33bc4bc514d4b","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"eval","value":541.0,"rawValue":541.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=IFEval - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-56454ddb45e74256015bb93c","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"ifeval-strict-acc","value":0.7837338262476897,"rawValue":0.7837338262476897,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=25;column=IFEval - IFEval Strict Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-cf4154f2ee245d43ce393d5f","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"observed-inference-time-s","value":33.75197721056489,"rawValue":33.75197721056489,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=25;column=IFEval - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1a50f1e54e9ff9ceef678381","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"output-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=IFEval - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-2fa5a2d521eda8fdff0d2c32","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"prompt-tokens","value":46.208872458410355,"rawValue":46.208872458410355,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=IFEval - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-aa5e6def12186254da0d39f3","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=IFEval - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-baee3615440a7b6adb7e5402","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=IFEval - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-5ade6afbfb7c844f821e88aa","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"eval","value":541.0,"rawValue":541.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=IFEval - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-9d71fcba6cbcc6db457b8e92","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"ifeval-strict-acc","value":0.874922982131855,"rawValue":0.874922982131855,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=55;column=IFEval - IFEval Strict Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b964f6dc1092a0d579c1cf57","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"observed-inference-time-s","value":35.937195608664354,"rawValue":35.937195608664354,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=55;column=IFEval - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-8f5887fcc50098747d386016","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"output-tokens","value":527.6414048059149,"rawValue":527.6414048059149,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=IFEval - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-8139d463558e30db9e6e9138","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"prompt-tokens","value":45.67097966728281,"rawValue":45.67097966728281,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=IFEval - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-62bbcb47ed1b1bfed7b4ea15","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=IFEval - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c62dc597cc16e4cd0249e00a","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=IFEval - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1809a2f5df9d4907b5943505","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"eval","value":541.0,"rawValue":541.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=IFEval - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-850d09e2057a6f2314c9edd8","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"ifeval-strict-acc","value":0.9269870609981515,"rawValue":0.9269870609981515,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=56;column=IFEval - IFEval Strict Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c400529f2dc9c5a2aeeee358","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"observed-inference-time-s","value":18.83414089833963,"rawValue":18.83414089833963,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=56;column=IFEval - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-6f98a25a5c1cf68b699bb5d8","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"output-tokens","value":441.13678373382623,"rawValue":441.13678373382623,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=IFEval - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-d56ae9a382109738cc5fabaa","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"prompt-tokens","value":45.67097966728281,"rawValue":45.67097966728281,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=IFEval - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-ae8b32c0aa016cd1233335db","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=IFEval - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-373a165927d45aa72591aff9","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=IFEval - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-6a0ca6909c3d103dfa83a0af","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"eval","value":541.0,"rawValue":541.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=IFEval - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-dcc1c565c45f43eb359426c5","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"ifeval-strict-acc","value":0.9319162045594579,"rawValue":0.9319162045594579,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=57;column=IFEval - IFEval Strict Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-cfe90f169154737c3043292c","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"observed-inference-time-s","value":26.734930773980075,"rawValue":26.734930773980075,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=57;column=IFEval - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-2e9063510905cfddbbac2098","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"output-tokens","value":426.6561922365989,"rawValue":426.6561922365989,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=IFEval - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1859b6b792ecd5d41ee36e84","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"prompt-tokens","value":45.67097966728281,"rawValue":45.67097966728281,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=IFEval - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-5177c0585233a841de8872ca","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=IFEval - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-28d114811a2e67f41d11b7ef","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=IFEval - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-88022f74374bdac31b39de57","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"eval","value":541.0,"rawValue":541.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=IFEval - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-59561b4573270c470f47025c","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"ifeval-strict-acc","value":0.9349969192852746,"rawValue":0.9349969192852746,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=58;column=IFEval - IFEval Strict Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d36ffa114c552942a44d7620","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"observed-inference-time-s","value":13.15882584436103,"rawValue":13.15882584436103,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=58;column=IFEval - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-44a55b4207de4b3e0ea8bda4","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"output-tokens","value":647.0628465804067,"rawValue":647.0628465804067,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=IFEval - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-2cafe5186cc0858b86956e90","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"prompt-tokens","value":45.67097966728281,"rawValue":45.67097966728281,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=IFEval - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-fd56e42363ea92c4625642b2","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=IFEval - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-bbada9a9b72b5cdba724207d","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-ifeval","benchmarkName":"IFEval · HELM Capabilities","benchmarkVersion":"v1.15.0","benchmarkVersionHint":"helm-ifeval@v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=IFEval - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-0744bb54252ae91fc703dcbc","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"efficiency-score","value":20.48127702555515,"rawValue":20.48127702555515,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=36;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"sourceMetricId":"score"},{"id":"pub-3b7428148249388f11afb0c0","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"score","value":0.7573661731997173,"rawValue":0.7573661731997173,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=36;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-8af517d3b58ff2f35f07be6c","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"efficiency-score","value":52.297304217949794,"rawValue":52.297304217949794,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=37;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"score"},{"id":"pub-4e56d4eced00ec918947be41","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"score","value":0.7800254913409985,"rawValue":0.7800254913409985,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=37;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8d78d9ab968f70e0293135aa","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"efficiency-score","value":7.381503096938465,"rawValue":7.381503096938465,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=39;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"score"},{"id":"pub-9e7332754f47cdb81f431db0","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"score","value":0.7167516798874939,"rawValue":0.7167516798874939,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=39;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a96097a99d87d4667995a83e","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"efficiency-score","value":17.536448448412127,"rawValue":17.536448448412127,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=38;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"score"},{"id":"pub-06d8696d6c867befbfc93566","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"score","value":0.7624408441066268,"rawValue":0.7624408441066268,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=38;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2c94a114023f7006664529ce","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"efficiency-score","value":115.28182297150872,"rawValue":115.28182297150872,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=25;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"sourceMetricId":"score"},{"id":"pub-a163c1ce249892d2865b14bf","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"score","value":0.6989491402703004,"rawValue":0.6989491402703004,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=25;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c1dc9926b50781ce4fdf76c1","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"efficiency-score","value":74.66990821942755,"rawValue":74.66990821942755,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=55;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"score"},{"id":"pub-24028321c200b6a2eb5e105c","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"score","value":0.806750930922304,"rawValue":0.806750930922304,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=55;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a032dcce84d68a02c8069cce","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"efficiency-score","value":28.206869066978612,"rawValue":28.206869066978612,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=56;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"score"},{"id":"pub-7448497435490e16f84e6134","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"score","value":0.8189777476920754,"rawValue":0.8189777476920754,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=56;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b71b9350508e753a88be92bb","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"efficiency-score","value":48.213836350621065,"rawValue":48.213836350621065,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=57;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"sourceMetricId":"score"},{"id":"pub-e9864a3857d42462be85fddd","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"score","value":0.7483391618552583,"rawValue":0.7483391618552583,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=57;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a607de0b4e8dada34313d893","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"efficiency-score","value":10.620566227529599,"rawValue":10.620566227529599,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=58;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"sourceMetricId":"score"},{"id":"pub-d20e3d2e031ffcf1cfb9883a","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-mean-score","benchmarkVersion":"v1.15.0","metricId":"score","value":0.656429079920834,"rawValue":0.656429079920834,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=58;column=Mean score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4dc1b4cde28b8e8d68fb2bc7","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.859,"rawValue":0.859,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=36;column=MMLU-Pro - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-62463aece6b6a11e06306368","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=MMLU-Pro - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-dec0ae012fef20af278256bf","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":12.630421590518665,"rawValue":12.630421590518665,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=36;column=MMLU-Pro - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a2e6aaf608e6af7848d91dfb","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":344.469,"rawValue":344.469,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=MMLU-Pro - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-d0cb144d91b5cca2f3b781f5","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":230.461,"rawValue":230.461,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=MMLU-Pro - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-323c532b00bee63083110874","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=MMLU-Pro - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-8a3dd521fa93b0156a10e0fe","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=MMLU-Pro - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-29b092e9ce38c30f03a0a717","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.875,"rawValue":0.875,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=37;column=MMLU-Pro - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-5bb38929fde4df70d9181d85","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=MMLU-Pro - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-b78864ce6a16548874675ac7","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":28.46593898815197,"rawValue":28.46593898815197,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=37;column=MMLU-Pro - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-75afd06b104964828da09508","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":272.871,"rawValue":272.871,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=MMLU-Pro - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3c0899e5401552537ded1fe8","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":252.461,"rawValue":252.461,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=MMLU-Pro - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-5aab4f6497d484e2120aab12","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=MMLU-Pro - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-655d113fdcf0e5bd6b52c7c3","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=MMLU-Pro - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-0cce23b25e106f53793294a9","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.777,"rawValue":0.777,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=39;column=MMLU-Pro - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b705b87ee3fce8ed31510038","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=MMLU-Pro - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-b2d87261d27f3b44efbbc748","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":3.7008020806312563,"rawValue":3.7008020806312563,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=39;column=MMLU-Pro - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-944b0b5de2efef1acdddccae","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":374.129,"rawValue":374.129,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=MMLU-Pro - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-964552be6d1b189d0f1ab2cb","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":252.461,"rawValue":252.461,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=MMLU-Pro - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a6a1986c54ef3f0e565bea38","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=MMLU-Pro - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c32e7b1257d82f1905b1bf5b","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=MMLU-Pro - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-28b85b02e2a048b002fe7473","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.869,"rawValue":0.869,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=38;column=MMLU-Pro - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6ef956645a616556cd151b4c","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=MMLU-Pro - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a583d4a278730509df2036cf","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":9.029817205530268,"rawValue":9.029817205530268,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=38;column=MMLU-Pro - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-bad6b08dba68336dc56dab47","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":392.292,"rawValue":392.292,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=MMLU-Pro - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3b8b4976704354b429043442","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":252.461,"rawValue":252.461,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=MMLU-Pro - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-cee85663dfe60953484b028e","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=MMLU-Pro - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-0ca0a247695a3a1d8140bdc1","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=MMLU-Pro - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-33666e1c9f1f868f19a8cf2f","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.793,"rawValue":0.793,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=25;column=MMLU-Pro - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a0bca293f040055e8d1dbf08","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=MMLU-Pro - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-82de2f2745d727cd8b849237","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":91.01470815229416,"rawValue":91.01470815229416,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=25;column=MMLU-Pro - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-363e3c1d2807a8f55f6de709","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=MMLU-Pro - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-cddf08deafd23875e988cf10","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":248.757,"rawValue":248.757,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=MMLU-Pro - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-8c9c8f49485895fb51e7630a","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=MMLU-Pro - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-148d22fe03b032d2beaf3be7","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=MMLU-Pro - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-10e62447707d9f3df009a411","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.863,"rawValue":0.863,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=55;column=MMLU-Pro - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d2aa47b70002312d379b1b46","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=MMLU-Pro - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-9c4568356dba7a5b1f335428","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":18.668269051074983,"rawValue":18.668269051074983,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=55;column=MMLU-Pro - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-772456a5e37d5e4c4d2191c4","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":5.028,"rawValue":5.028,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=MMLU-Pro - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-de75be1bce3657c2fbbea246","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":248.569,"rawValue":248.569,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=MMLU-Pro - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c6c3e6817bf9c8428d52286d","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=MMLU-Pro - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-9050ced713f303c28ab09883","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=MMLU-Pro - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-799ea9bda61c664b9f91a0dd","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.835,"rawValue":0.835,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=56;column=MMLU-Pro - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e428e6903c580502cceffcfb","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=MMLU-Pro - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-afceb5ab1850eb4c52976bf0","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":11.802515007257462,"rawValue":11.802515007257462,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=56;column=MMLU-Pro - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1845f4467c1af6c2a692d213","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":17.495,"rawValue":17.495,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=MMLU-Pro - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-e06c90d4482364cf13ff8b06","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":248.569,"rawValue":248.569,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=MMLU-Pro - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-924caa79923cb31fe5bae007","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=MMLU-Pro - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-7a368fab015bbf6490381e20","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=MMLU-Pro - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-699bd1d9ce3dfee555ea2566","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.778,"rawValue":0.778,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=57;column=MMLU-Pro - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7427dad156fda777a7a96427","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=MMLU-Pro - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-7355f22bbcec5cf659aad4a6","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":17.336622306585312,"rawValue":17.336622306585312,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=57;column=MMLU-Pro - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-86ad5496365142140e649e09","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":5.385,"rawValue":5.385,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=MMLU-Pro - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-eaec22cdfdd9b7bbdec893c5","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":248.569,"rawValue":248.569,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=MMLU-Pro - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-fe43ee796f849b5f5a37bcc7","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=MMLU-Pro - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-ec043a7f3eea1991e31d2fdc","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=MMLU-Pro - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-aad1b27d6869cf445d4c0689","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"cot-correct","value":0.579,"rawValue":0.579,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=58;column=MMLU-Pro - COT correct","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d4700a3b45db4641e1a326b5","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=MMLU-Pro - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-99390f292e52c2eadf1a4cbf","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":1.1470122172832489,"rawValue":1.1470122172832489,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=58;column=MMLU-Pro - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a16516a3fea25694e2b5b038","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":5.002,"rawValue":5.002,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=MMLU-Pro - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-8dd7c1139fc959d8cb88289f","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":248.569,"rawValue":248.569,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=MMLU-Pro - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-b7619b934c250742f7a9e84c","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=MMLU-Pro - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-6cb5c1a4e608d9202ec3d10b","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-mmlu-pro","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=MMLU-Pro - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-31662e4d900d8c1a52e33ffa","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"acc","value":0.5111666666666665,"rawValue":0.5111666666666665,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=36;column=Omni-MATH - Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-66b83c8b54e04588161bfc6f","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=Omni-MATH - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-10505c08c9f43b98544806ab","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":27.025822179079057,"rawValue":27.025822179079057,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=36;column=Omni-MATH - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-48d5c460ee7dac4d52c0ac5e","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":893.894,"rawValue":893.894,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=Omni-MATH - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-21dd46890ff147e8e6a16bba","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":110.563,"rawValue":110.563,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=Omni-MATH - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c441237dd23706e9e2bb5954","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=Omni-MATH - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-117976cb7e525365b7b10e1f","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=Omni-MATH - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-69cdcf08bd40ebb7dc598af3","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"acc","value":0.6160000000000001,"rawValue":0.6160000000000001,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=37;column=Omni-MATH - Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-77504be8cdfaadbfc72bb5c0","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=Omni-MATH - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-cd7119a265dc76226677de61","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":114.84836924410313,"rawValue":114.84836924410313,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=37;column=Omni-MATH - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-24d0e101d3b4e9265d45ea93","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":691.066,"rawValue":691.066,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=Omni-MATH - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-72027dafcd1b2b77db6fccf1","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":110.563,"rawValue":110.563,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=Omni-MATH - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-2ea4f1df14cf638913dea37d","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=Omni-MATH - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-9ef46f0e16eec205e4a59ae5","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=Omni-MATH - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1561a5303812a8bf11ebac16","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"acc","value":0.561,"rawValue":0.561,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=39;column=Omni-MATH - Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-875ba97c748e3d9f4d8e5eb3","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=Omni-MATH - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a70f394d7910fc54ca8be1ff","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":7.431977860689163,"rawValue":7.431977860689163,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=39;column=Omni-MATH - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-5408e32576fa045a8f11758e","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":937.799,"rawValue":937.799,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=Omni-MATH - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-422c066cce08a665dbd0aa3c","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":110.563,"rawValue":110.563,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=Omni-MATH - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-cb292d1e255b7b4bb303d70f","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=Omni-MATH - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-346e1efd2a30bcec8bd7d892","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=Omni-MATH - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-d25fedfc3955d1313d1de2d0","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"acc","value":0.5529999999999999,"rawValue":0.5529999999999999,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=38;column=Omni-MATH - Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-519ed1d141faa9c375e8a095","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=Omni-MATH - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-69d00d4d8792b8298c611023","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":16.790386716127397,"rawValue":16.790386716127397,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=38;column=Omni-MATH - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3fa658a3332268b9eb35747b","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":892.774,"rawValue":892.774,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=Omni-MATH - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-183d0c508dd9a998deb00939","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":110.563,"rawValue":110.563,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=Omni-MATH - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-2436f0769bd2883f88f9f966","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=Omni-MATH - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-d08d9ef2f8e1d8b598d83f51","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=Omni-MATH - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-13ecfc101c0c14fa346f1360","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"acc","value":0.42400000000000004,"rawValue":0.42400000000000004,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=25;column=Omni-MATH - Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-414846420fa1d41de7d49b7c","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=Omni-MATH - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-d2f005a1c9b325e9429522d5","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":208.35547973060608,"rawValue":208.35547973060608,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=25;column=Omni-MATH - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-9b893874dce851cc5a381a63","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=Omni-MATH - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c1faf3f22855bf4ae684b8d9","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":107.102,"rawValue":107.102,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=Omni-MATH - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-cba52e1facc7e7079c53a1f2","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=Omni-MATH - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-06c98d8815a52459b90db745","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=Omni-MATH - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c2bd026fd547d3b2e9832fc9","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"acc","value":0.647,"rawValue":0.647,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=55;column=Omni-MATH - Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-86072d43c0dace2ba8c9091b","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=Omni-MATH - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-74737fa1b0410fc97cd61ca4","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":172.73094402194022,"rawValue":172.73094402194022,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=55;column=Omni-MATH - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-2e2198a8496a8ad40eab5afe","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":228.774,"rawValue":228.774,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=Omni-MATH - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-9423242d972a94d57ff043de","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":109.623,"rawValue":109.623,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=Omni-MATH - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-f30bf51fa79d3d77b5ee4cc1","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=Omni-MATH - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-45aca89a2eefbbabfebcd5ac","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=Omni-MATH - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1107fd367e7c964aae4f7417","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"acc","value":0.7220000000000001,"rawValue":0.7220000000000001,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=56;column=Omni-MATH - Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1cbcd2e4a051908e985935a7","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=Omni-MATH - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-9e39d39ca2e640ae5a59185d","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":55.806421615123746,"rawValue":55.806421615123746,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=56;column=Omni-MATH - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-f371f5236e465873ae63d91c","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":362.654,"rawValue":362.654,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=Omni-MATH - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a41284a985a12894b76ac610","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":109.623,"rawValue":109.623,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=Omni-MATH - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-b2bc9d4217a8072b7a36e131","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=Omni-MATH - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-903f2ce4fba5797237a2d119","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=Omni-MATH - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-735c2c396def5db2b87d3a15","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"acc","value":0.5465000000000001,"rawValue":0.5465000000000001,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=57;column=Omni-MATH - Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2dfad70de6362491e8e0277e","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=Omni-MATH - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3bec7de6fa401b29f519bc0f","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":119.19145288252831,"rawValue":119.19145288252831,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=57;column=Omni-MATH - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-2d5f35b5b2592cbcb8037cd0","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":222.15,"rawValue":222.15,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=Omni-MATH - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-e42e0eaec3dcdefd00bbe931","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":109.623,"rawValue":109.623,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=Omni-MATH - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-0304d441e12602650eb7fa62","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=Omni-MATH - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-22183192965a46c0a1993e4c","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=Omni-MATH - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-4ec707e7a685f5e56a9160df","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"acc","value":0.4635,"rawValue":0.4635,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=58;column=Omni-MATH - Acc","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-967b97b7819c680b538c61e6","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=Omni-MATH - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-4c64b2829d7b757bee94c930","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":9.713221177577973,"rawValue":9.713221177577973,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=58;column=Omni-MATH - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-8dab701e7c16bc03aeeacc53","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":1256.266,"rawValue":1256.266,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=Omni-MATH - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3b1dc0aabcee6333f6cda88b","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":109.623,"rawValue":109.623,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=Omni-MATH - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-10886e90a22f028b3375b504","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=Omni-MATH - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-eaed9d1e52fa6cbbeb517c7a","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-omni-math","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=Omni-MATH - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c78f0fc943e44b907a821343","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=WildBench - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3ab34fc5554f75b5ae31698f","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":29.848318881988526,"rawValue":29.848318881988526,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=36;column=WildBench - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-2994752cf2d410c24d71acd0","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":936.927,"rawValue":936.927,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=WildBench - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-6df75c17d8a60c9865882b70","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=WildBench - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-0cb9ad9b268d8315e1fbaa8f","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=WildBench - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3837237220087265149c4371","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=36;column=WildBench - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-41aaa3bc3e91b6399ccdea3e","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"wb-score","value":0.833000000000007,"rawValue":0.833000000000007,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=36;column=WildBench - WB Score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-fe23a2c6ce43111560fdbb39","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=WildBench - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-8029c5a8992e11f81538f203","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":50.19046350765228,"rawValue":50.19046350765228,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=37;column=WildBench - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-cdb369a3a411a40135285d19","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":1195.769,"rawValue":1195.769,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=WildBench - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-416a5ad67aa7b0d977e60523","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=WildBench - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-bb6b1095638070ebaf2be889","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=WildBench - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-7c9708b75337ee512fec0e9e","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=37;column=WildBench - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-fe794580cd60f4cbb378edcb","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus (20250514, extended thinking)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"wb-score","value":0.8518703703703797,"rawValue":0.8518703703703797,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=37;column=WildBench - WB Score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d8ceb100258db933704ed0ab","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=WildBench - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-e7e30598dd7621c8190a12b5","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":16.317131044387818,"rawValue":16.317131044387818,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=39;column=WildBench - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-fb9c98c8f055c1e29d4a7edb","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":1835.337,"rawValue":1835.337,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=WildBench - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-ec1e9933b7ec6965f274ec63","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=WildBench - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-97489ef39730547c2dffe7f8","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=WildBench - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-abffd643db0d91bfd1b01d44","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=39;column=WildBench - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-0da8006ffa1e5ddd8e2fcec0","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku (20251001)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"wb-score","value":0.8390833333333435,"rawValue":0.8390833333333435,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=39;column=WildBench - WB Score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9c678bffa01049fc61e41a20","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=WildBench - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-11c70a0e094768c442658f62","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":38.54364204096484,"rawValue":38.54364204096484,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=38;column=WildBench - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-6acf50b7b4ca2eb4119066d0","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":1804.604,"rawValue":1804.604,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=WildBench - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-f02b705c59c7ce376eebcbed","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=WildBench - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-80737d933d3bef5d085a619f","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=WildBench - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-90c9c451a401216c5da03bbf","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=38;column=WildBench - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-903b914a0bbc990b64eb61b4","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet (20250929)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"wb-score","value":0.8544444444444534,"rawValue":0.8544444444444534,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=38;column=WildBench - WB Score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-81bfea7d37f030899f55db3d","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=WildBench - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-ae1f8918a6e0b131ae0c6a52","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":87.84843708276749,"rawValue":87.84843708276749,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=25;column=WildBench - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-15613943690143fe02a7039a","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=WildBench - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-61a8c7bf1a107cf8ce104f89","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=WildBench - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-b0346c2d6d2fc2369b092ec9","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=WildBench - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-ce41093fad63a1ce1c342312","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=25;column=WildBench - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-4f25c3d88b94d2767cd2ce1d","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"wb-score","value":0.8280925925926018,"rawValue":0.8280925925926018,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=25;column=WildBench - WB Score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0a41b8d6fc502331befe7d48","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=WildBench - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-6f98a3cd87d62fa572260f3d","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":88.59490567517281,"rawValue":88.59490567517281,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=55;column=WildBench - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a049bf4dde61407eb8b85f7a","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":1518.974,"rawValue":1518.974,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=WildBench - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-b9b0ba6715f62c6bfcf18adc","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=WildBench - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-5c25cb61e832388c667d0dcc","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=WildBench - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c3e75a81d0fc6e076f5fac81","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=55;column=WildBench - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-af0f5da55fd755f4e42931b6","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5 (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"wb-score","value":0.8573518518518621,"rawValue":0.8573518518518621,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=55;column=WildBench - WB Score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-07b78eef4757d87bc213061a","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=WildBench - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-12fd94cdd5cccafa12683d74","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":33.85394237089157,"rawValue":33.85394237089157,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=56;column=WildBench - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-a8106697f39f08f1cdb8d427","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":1408.024,"rawValue":1408.024,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=WildBench - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-c5f4a74cb43992315e72fe9d","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=WildBench - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-39c725b030e2ec3bc3b7d55e","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=WildBench - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-0e44844d778423e1bafad80c","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=56;column=WildBench - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-ab355b0db1b38766c085e73f","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT-5 mini (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"wb-score","value":0.855296296296306,"rawValue":0.855296296296306,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=56;column=WildBench - WB Score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-74e1ad2a9a64064397e7c5a8","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=WildBench - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-2eebfd77e62ea86024f00fc8","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":47.560468022584914,"rawValue":47.560468022584914,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=57;column=WildBench - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-23132c3f213d935ceb540988","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":1391.144,"rawValue":1391.144,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=WildBench - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-1145ae93415c18a5fe74c232","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=WildBench - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-8e380fcc6f8e5052fc18a557","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=WildBench - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-3cdcebf6ea1f2d850fcc2385","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=57;column=WildBench - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-958e5cd3e86108a1de7c50ba","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT-5 nano (2025-08-07)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"wb-score","value":0.8059074074074162,"rawValue":0.8059074074074162,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=57;column=WildBench - WB Score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-23a7cc01cbd510c178261925","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"eval","value":1000.0,"rawValue":1000.0,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=WildBench - # eval","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-d3bb9d4352481476b10d25c9","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"observed-inference-time-s","value":28.08133857488632,"rawValue":28.08133857488632,"unit":"seconds","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Efficiency;row=58;column=WildBench - Observed inference time (s)","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Efficiency"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-4d6b94988ac221e609252f96","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"output-tokens","value":2059.716,"rawValue":2059.716,"unit":"count","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=WildBench - # output tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-e4e8a7be64bebaab43d62e2f","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"prompt-tokens","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=WildBench - # prompt tokens","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-f9201e418cf2b076c1431d0f","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"train","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=WildBench - # train","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"lower","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-4588cb7abe23a30456032d7d","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"truncated","value":0.0,"rawValue":0.0,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=General information;row=58;column=WildBench - truncated","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"General information"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"matrixExcluded":true,"matrixExcludedReason":"telemetry_metric"},{"id":"pub-953a4e2cbdc540707460fcf1","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT-5.1 (2025-11-13)","benchmarkId":"helm-wildbench","benchmarkVersion":"v1.15.0","metricId":"wb-score","value":0.8629444444444558,"rawValue":0.8629444444444558,"unit":"fraction","evidenceLevel":"A","sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourcePageUrl":"https://crfm.stanford.edu/helm/capabilities/latest/config.js","sourceLocator":"release=v1.15.0;table=Accuracy;row=58;column=WildBench - WB Score","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2025-11-24","payloadSha256":"63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","protocol":{"direction":"higher","harness":"helm","project":"capabilities","release":"v1.15.0","source_status":"published_release","table":"Accuracy"},"harnessId":"helm","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cbc0d8e327c2215ef8546c4a","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"MiniMaxAI/MiniMax-M2.1","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":22.2,"rawValue":22.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=58;row=57","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M2.1"},{"id":"pub-86dfdca7e71d74cc14b16f12","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMaxAI/MiniMax-M2.5","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":19.4,"rawValue":19.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=64;row=63","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M2.5"},{"id":"pub-0fbcbaa0d06847989ad31990","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"Qwen/Qwen3.5-122B-A10B","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":47.5,"rawValue":47.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=17;row=16","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle_with_tools_4.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-122B-A10B"},{"id":"pub-4c0e8f9762c828a6a5c00232","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"Qwen/Qwen3.5-122B-A10B","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":25.3,"rawValue":25.3,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=45;row=44","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle_chain_of_thought_3.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-122B-A10B"},{"id":"pub-edf4a2019a2785baf73cd3a4","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"Qwen/Qwen3.5-27B","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":48.5,"rawValue":48.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=12;row=11","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle_with_tools_3.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-27B"},{"id":"pub-57b97116f5fae7ff54a8f7ed","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"Qwen/Qwen3.5-27B","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":24.3,"rawValue":24.3,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=48;row=47","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle_chain_of_thought_2.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-27B"},{"id":"pub-a078e0829bd60ad12ee0d6dc","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"Qwen/Qwen3.5-397B-A17B","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":48.3,"rawValue":48.3,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=13;row=12","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle_with_tools.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-397B-A17B"},{"id":"pub-aec354ba47bfab112a7b1f11","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"Qwen/Qwen3.5-397B-A17B","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":28.7,"rawValue":28.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=40;row=39","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-397B-A17B"},{"id":"pub-62e0f19528bf40bf93428b2a","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"Qwen/Qwen3.6-27B","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":24.0,"rawValue":24,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=49;row=48","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-27B"},{"id":"pub-3938dfac52cf6baa0eef0f83","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"Qwen/Qwen3.6-35B-A3B","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":21.4,"rawValue":21.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=62;row=61","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-35B-A3B"},{"id":"pub-e0521489452a41891053df36","canonicalModelId":"qwen/qwen3.8-2.4t-a95b@2026-08-12","modelName":"Qwen3.8 2.4T A95B","modelRef":"Qwen/Qwen3.8-2.4T-A95B","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":43.6,"rawValue":43.6,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=21;row=20","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B"},{"id":"pub-f1079df2a35bdd2c40e1655b","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"Qwen/Qwen3.8-27B","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":30.8,"rawValue":30.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=35;row=34","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/Qwen3.8-27B.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.8-27B"},{"id":"pub-88d9ed2df90763666f3c0c43","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"Qwen/Qwen3.8-Flash-Next","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":35.9,"rawValue":35.9,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=27;row=26","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/Qwen3.8-Flash-Next.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.8-Flash-Next"},{"id":"pub-d8782ae0c844eddcf649195a","canonicalModelId":"xiaomi/mimo-v2-flash@current","modelName":"MiMo V2 Flash","modelRef":"XiaomiMiMo/MiMo-V2-Flash","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":22.1,"rawValue":22.1,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=59;row=58","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/XiaomiMiMo/MiMo-V2-Flash"},{"id":"pub-08bf7297e416b9cb93a01a21","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"XiaomiMiMo/MiMo-V2.5-Pro","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":48.0,"rawValue":48,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=15;row=14","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro"},{"id":"pub-cc45ce0825586adb23027cee","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":6.68,"rawValue":"0.0668","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=41;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b698653ee5225f22d38ea160","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":10.72,"rawValue":"0.1072","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=30;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-3a41e0e83caf8f078955759d","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":14.16,"rawValue":"0.1416","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=25;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-36b7aa176d1a808b1875ea1a","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":25.2,"rawValue":"0.252","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=13;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-eda40c62418b192d9a35a262","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":19.0,"rawValue":"0.19","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=20;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-623af3eab03cc59f08d007c5","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":34.44,"rawValue":"0.3444","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=8;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ddd92195cceff27a61f5321c","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":36.199999999999996,"rawValue":"0.362","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=7;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4f216dfade4ce70f7bf74bf7","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":7.5200000000000005,"rawValue":"0.0752","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=39;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1ab8a110fcd273f0a73db7d2","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":13.720000000000002,"rawValue":"0.13720000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=26;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-58c2f9c488693fc8b274ddf7","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-ai/DeepSeek-V4-Flash","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":34.8,"rawValue":34.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=28;row=27","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash"},{"id":"pub-679f7fd725c79a4d61795f48","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-ai/DeepSeek-V4-Pro","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":37.7,"rawValue":37.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=24;row=23","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro"},{"id":"pub-dc438fc6e54129fd392825d0","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2026-02-19","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":46.44,"rawValue":"0.4644","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=2;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-5a6f1b489d59edc2fb3e5cca","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":25.319999999999997,"rawValue":"0.2532","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=11;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-e396bf7aaff94421b973a017","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":25.319999999999997,"rawValue":"0.2532","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=12;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-17bbe28a2c5a1e13a89ebff5","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":19.44,"rawValue":"0.19440000000000002","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=18;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-aea6ffa429124373d2868e09","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_none","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":6.800000000000001,"rawValue":"0.068","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=40;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ffaa5d14b500cc5592dac91c","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":23.68,"rawValue":"0.2368","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=15;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4c9f36038c51123d3fcaa13d","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":27.800000000000004,"rawValue":"0.278","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=10;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-879d4f51aa237bbf985b3f1f","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":36.24,"rawValue":"0.3624","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=6;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-92080b202e442249a41f2d68","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_unknown","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":44.32,"rawValue":"0.4432","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=3;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fde97f71b881ad4688a9a0a1","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersion":"2026-01-27","benchmarkVersionHint":"hle@v1","metricId":"Accuracy","value":24.370000000000005,"rawValue":"0.24370000000000003","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"hle_external.csv:row=14;column=Accuracy","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9191907b3e8262ab7dc6f92e","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"moonshotai/Kimi-K2.5","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":50.2,"rawValue":50.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=11;row=10","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle_with_tools.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.5"},{"id":"pub-c7578c808012322d95ee7aff","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"moonshotai/Kimi-K2.5","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":24.37,"rawValue":24.37,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=47;row=46","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/aa/05/aa055fe5-5174-4517-8b10-19d46da94f3e.json"},{"id":"pub-d2793a98cb3d2c41f6c62f46","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"moonshotai/Kimi-K2.6","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":54.0,"rawValue":54,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=5;row=4","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle_with_tools.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6"},{"id":"pub-0f6006e3ecd91151e70d8e5c","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"moonshotai/Kimi-K2.6","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":34.7,"rawValue":34.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=29;row=28","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6"},{"id":"pub-98af02799329368f475b73df","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"moonshotai/Kimi-K3","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":56.0,"rawValue":56,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=2;row=1","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K3"},{"id":"pub-a9612a5c3c5c046442a41aca","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/GLM-4.7","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":42.8,"rawValue":42.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=22;row=21","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle_with_tools.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-4.7"},{"id":"pub-177ba21f61bdec464631184e","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/GLM-4.7","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":24.8,"rawValue":24.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=46;row=45","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/zai-org/GLM-4.7"},{"id":"pub-0d0d4e783b767dd6da2389b1","canonicalModelId":"zai/glm-4.7-flash@current","modelName":"GLM-4.7 Flash","modelRef":"zai-org/GLM-4.7-Flash","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":14.4,"rawValue":14.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=73;row=72","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-4.7-Flash"},{"id":"pub-b484de33869a84985d88e436","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"zai-org/GLM-5","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":50.4,"rawValue":50.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=10;row=9","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle_with_tools.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5"},{"id":"pub-39c6d296a205a0a057b7c5e3","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"zai-org/GLM-5","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":30.5,"rawValue":30.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=36;row=35","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5"},{"id":"pub-ac845176305cd06ce4019def","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"zai-org/GLM-5.1","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":52.3,"rawValue":52.3,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=9;row=8","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle_with_tools.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.1"},{"id":"pub-f08479ac5efda2747ccb4a42","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"zai-org/GLM-5.1","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":31.0,"rawValue":31,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=34;row=33","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.1"},{"id":"pub-d9d97dcd4af4f1d6e37ed767","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"zai-org/GLM-5.2","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":54.7,"rawValue":54.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=4;row=3","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.2"},{"id":"pub-4476e035caf9caea483a1d94","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"zai-org/GLM-5.2","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":40.5,"rawValue":40.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=23;row=22","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/hle.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.2"},{"id":"pub-bda8d7fd370b92bdb71eae4c","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"zai-org/GLM-5.3-Flash","benchmarkId":"hle","benchmarkName":"Humanity's Last Exam","benchmarkVersionHint":"hle@v1","metricId":"score","value":55.3,"rawValue":55.3,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","sourceLocator":"rank=3;row=2","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","protocol":{"filename":".eval_results/GLM-5.3-Flash.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.3-Flash"},{"id":"pub-6bc206685f5c7bf531b80703","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"Qwen/Qwen3.5-27B","benchmarkId":"hmmt","benchmarkName":"HMMT","benchmarkVersionHint":"hmmt@2026","metricId":"score","value":81.06,"rawValue":81.06,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourceLocator":"rank=14;row=13","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2e159d9614de10ac2db4a6f240abde1aa936df4b4a22fe613d61729786f322b9","protocol":{"filename":".eval_results/MathArena--hmmt_feb_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://matharena.ai/?comp=hmmt--hmmt_feb_2026"},{"id":"pub-adc4d0c7b4ce08c78e6e1ba8","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"Qwen/Qwen3.5-397B-A17B","benchmarkId":"hmmt","benchmarkName":"HMMT","benchmarkVersionHint":"hmmt@2026","metricId":"score","value":87.88,"rawValue":87.88,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourceLocator":"rank=4;row=3","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2e159d9614de10ac2db4a6f240abde1aa936df4b4a22fe613d61729786f322b9","protocol":{"filename":".eval_results/MathArena--hmmt_feb_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://matharena.ai/?comp=hmmt--hmmt_feb_2026"},{"id":"pub-35ff96473af6acd55251dfca","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"Qwen/Qwen3.6-27B","benchmarkId":"hmmt","benchmarkName":"HMMT","benchmarkVersionHint":"hmmt@2026","metricId":"score","value":84.3,"rawValue":84.3,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourceLocator":"rank=11;row=10","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2e159d9614de10ac2db4a6f240abde1aa936df4b4a22fe613d61729786f322b9","protocol":{"filename":".eval_results/hmmt_feb_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-27B"},{"id":"pub-879e3f654105211d00ac0c1f","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"Qwen/Qwen3.6-35B-A3B","benchmarkId":"hmmt","benchmarkName":"HMMT","benchmarkVersionHint":"hmmt@2026","metricId":"score","value":83.6,"rawValue":83.6,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourceLocator":"rank=12;row=11","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2e159d9614de10ac2db4a6f240abde1aa936df4b4a22fe613d61729786f322b9","protocol":{"filename":".eval_results/hmmt_feb_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-35B-A3B"},{"id":"pub-3c585259627eb607b7f52d8d","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"moonshotai/Kimi-K2.5","benchmarkId":"hmmt","benchmarkName":"HMMT","benchmarkVersionHint":"hmmt@2026","metricId":"score","value":87.12,"rawValue":87.12,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourceLocator":"rank=6;row=5","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2e159d9614de10ac2db4a6f240abde1aa936df4b4a22fe613d61729786f322b9","protocol":{"filename":".eval_results/MathArena--hmmt_feb_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://matharena.ai/?comp=hmmt--hmmt_feb_2026"},{"id":"pub-4130b33c5b0429795d94d747","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"moonshotai/Kimi-K2.6","benchmarkId":"hmmt","benchmarkName":"HMMT","benchmarkVersionHint":"hmmt@2026","metricId":"score","value":92.7,"rawValue":92.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourceLocator":"rank=1;row=0","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2e159d9614de10ac2db4a6f240abde1aa936df4b4a22fe613d61729786f322b9","protocol":{"filename":".eval_results/hmmt_feb_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6"},{"id":"pub-6d795dd58b1a8914e9497d07","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"zai-org/GLM-5","benchmarkId":"hmmt","benchmarkName":"HMMT","benchmarkVersionHint":"hmmt@2026","metricId":"score","value":86.36,"rawValue":86.36,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourceLocator":"rank=9;row=8","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2e159d9614de10ac2db4a6f240abde1aa936df4b4a22fe613d61729786f322b9","protocol":{"filename":".eval_results/MathArena--hmmt_feb_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://matharena.ai/?comp=hmmt--hmmt_feb_2026"},{"id":"pub-c5c7f45fac43f622863cca57","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"zai-org/GLM-5.1","benchmarkId":"hmmt","benchmarkName":"HMMT","benchmarkVersionHint":"hmmt@2026","metricId":"score","value":82.6,"rawValue":82.6,"unit":"percent","evidenceLevel":"C","sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","sourceLocator":"rank=13;row=12","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"2e159d9614de10ac2db4a6f240abde1aa936df4b4a22fe613d61729786f322b9","protocol":{"filename":".eval_results/MathArena--hmmt_feb_2026.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.1"},{"id":"pub-671f771e1a662a547f4ffbec","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"livebench","benchmarkName":"LiveBench","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"livebench@rolling","metricId":"Global average","value":78.79,"rawValue":"78.79","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"live_bench_external.csv:row=65;column=Global average","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://livebench.ai/#/"},{"id":"pub-fe0c7b0f25a682dda0a61d8f","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":99.0,"rawValue":"99.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d30cfee397ac8dcfb3cb6b82","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":99.0,"rawValue":"99.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d7ad543ea4026e95955f9396","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0f2bb829d3ef5531a8267d28","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0fd86807638348584211091a","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-84448e42a8ce27866e80f19f","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":99.01,"rawValue":"99.01","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8186553f7fc718059efd5a76","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":76.0,"rawValue":"76.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-19dc94f6fae6d881f38d82de","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f5d56b886a54dc5ab3580d82","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b57cf9ddcf4da0c4fd376c04","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-21983166716d3994cfc832a0","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":88.0,"rawValue":"88.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ec688a02af9e8cb56be90a6a","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f46c3c8b8d336816aa8d474f","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-50b09b69a69fdb380e83df08","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fe4c40feb79ddf8406e6e6d1","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d6010ff95d0270bbec6f9573","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":95.0,"rawValue":"95.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5e84c5ee11e477666e0d6f52","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a5c753fa8d725b0eaba6dd56","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9864b0e8a42558a99ef73c8a","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f50d17661bea3e4a2c02bab0","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":99.0,"rawValue":"99.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fbd66e4d79585c998a4051aa","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-02ddfeaf743ce4b75817a8d1","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-54836708455af060a2eca601","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-16e11253a7b4d134b3bdbb53","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-17749dd2aba21d80d8643ccb","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-25a37595cbfa6db5967b54b0","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-63ba99084651cc0fa8b430d5","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9fe3fe17b82dd04e7e475404","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-59443cd2e4c848b2908c8170","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-71072897943593d9469ca617","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0a60b164fa1ab551270245b3","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":99.0,"rawValue":"99.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-57c7de734667216931b801d9","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ed19b98c8f82fa3fe5ab7265","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-aafb638b8aa95e485c8432db","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4e9164d884eae426c9e0046b","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-53c6e1eccbfad3ccc3e9f39b","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":75.0,"rawValue":"75.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d203a41f91858d704080043f","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":93.0,"rawValue":"93.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-03d040180cc32144d034d225","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b83652be7babd625afe50186","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b6614c8a66c0e383ca00e918","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-32cdeeadc1963e7ec0a33bd7","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":99.0,"rawValue":"99.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8dcda60d8f0f5c9aceec395a","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-amps-hard","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=AMPS_Hard","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"AMPS_Hard"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-db4f07708e838a2fb9d6f680","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":80.435,"rawValue":"80.435","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c566a313b03e38087d7d9812","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":80.435,"rawValue":"80.435","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1e1fb13c437cfe17bd5535ce","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":76.087,"rawValue":"76.087","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-525387b122a6d83656ac6cfd","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":78.261,"rawValue":"78.261","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-13f33163c6aa6650388b5272","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":84.783,"rawValue":"84.783","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-670e9a180fe85f1a8f843c69","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":82.609,"rawValue":"82.609","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-19ea559be38d14af1a9fcc34","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":78.261,"rawValue":"78.261","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-57667621e01da443858f4be9","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":78.261,"rawValue":"78.261","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-55d6726206bf455542d98bed","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":65.217,"rawValue":"65.217","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-9e43894416c2ad2f3de362ef","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":73.913,"rawValue":"73.913","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1b2955186b8503bbb09ee6e0","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":67.391,"rawValue":"67.391","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5361e54757a9a17fb0f6aa00","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":69.565,"rawValue":"69.565","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d7c699041b3a66caca16922c","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":78.261,"rawValue":"78.261","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1c023c5b3848f434fa85042c","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":78.261,"rawValue":"78.261","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9bc1e88af7424c601f0ed5db","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":76.087,"rawValue":"76.087","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ade1a17605018e07ea6cc549","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":76.087,"rawValue":"76.087","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b73292427bff99a1ae6a0dea","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":78.261,"rawValue":"78.261","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e50206d75bb12509aa74d4ac","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":76.087,"rawValue":"76.087","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e868a3a5d65bae8326d8ef47","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":80.435,"rawValue":"80.435","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7a4302e4def7e7133ff1914a","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":80.435,"rawValue":"80.435","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7e9f0289888d44a604991dca","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":80.435,"rawValue":"80.435","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-74e807fa7a7a727e10b85bce","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":76.087,"rawValue":"76.087","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1e12478228fd1b2101b6959b","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":86.957,"rawValue":"86.957","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-97a4e64fcb592719476d09e9","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":75.556,"rawValue":"75.556","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7721d29544e8285b3ed4e541","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":67.391,"rawValue":"67.391","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a87e1fd4c84b824b65e2973f","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":80.435,"rawValue":"80.435","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-693be8360cd801e5c5849159","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":82.609,"rawValue":"82.609","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2809a0d0a6e5b0b6e07c1347","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":86.957,"rawValue":"86.957","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-407657160f70f091db686ee4","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":84.783,"rawValue":"84.783","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-818d4679b5b6f54d55a4ab7e","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":80.435,"rawValue":"80.435","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4d6def8e9564d0c0b2f92866","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":69.565,"rawValue":"69.565","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6694c5c628b8f8fca6d071a6","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":76.087,"rawValue":"76.087","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-63ef6a1f54d10805aceae688","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":78.261,"rawValue":"78.261","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ae4677b451332091006b5316","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":76.087,"rawValue":"76.087","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2dc64b7fb55db29cb3c82e4d","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":82.609,"rawValue":"82.609","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a6983a4a8816ef8b00bfc20f","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":67.391,"rawValue":"67.391","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ff166da40ce5113ee7b276ac","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":71.739,"rawValue":"71.739","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ac6e244613438b8cb9186faf","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":76.087,"rawValue":"76.087","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f70c68df0b2b84d46902e955","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":69.565,"rawValue":"69.565","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2c8fc988f4b1a4e7451140ea","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":73.913,"rawValue":"73.913","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-43bf7f2c5b9592b31460fbc8","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":76.087,"rawValue":"76.087","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-bced40a0ab472e25c0b89801","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-code-completion","benchmarkVersion":"2026_06_25","metricId":"score","value":73.913,"rawValue":"73.913","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=code_completion","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_completion"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e1427f074185de5df30ae05a","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":91.549,"rawValue":"91.549","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ebf81def3df1e7816ef4a345","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":78.873,"rawValue":"78.873","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b58b673514e680eb4bef3cfc","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.282,"rawValue":"80.282","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0d235c18737ed5f3df53b2a5","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":85.915,"rawValue":"85.915","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2c33ba50c5f89dea55133af8","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":78.873,"rawValue":"78.873","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6bd5e4feb327faa93afb9bf1","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.282,"rawValue":"80.282","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-aba0531146fbd6deeed7b288","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.282,"rawValue":"80.282","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3b98cf76dd8749a681461482","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":83.099,"rawValue":"83.099","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f118f6207fa3fb44a2c58d0f","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":73.239,"rawValue":"73.239","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-862b84db833c0da21febeec0","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":76.056,"rawValue":"76.056","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-b10cd2c58400688d9a6cfeb3","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":69.014,"rawValue":"69.014","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a94284400271436a5a772c9a","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":70.423,"rawValue":"70.423","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-05a45ea9110d57a46420f026","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":76.056,"rawValue":"76.056","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2dfb7467f2889008f645ab6b","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":74.648,"rawValue":"74.648","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-254f8cf1da07739d0553efbf","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.282,"rawValue":"80.282","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-92829d3f7ff9db77536832ca","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":76.056,"rawValue":"76.056","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e71b3ef5996cf04b52d0e502","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":77.465,"rawValue":"77.465","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e6f1c7a712c94d21aa4055c1","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":81.69,"rawValue":"81.69","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d1cd55da8476bc7a821980a4","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":78.873,"rawValue":"78.873","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4e28f68b1391d10329bbb3fa","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":77.465,"rawValue":"77.465","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d7002b941cf72ceadc4b21c8","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":77.465,"rawValue":"77.465","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ab0bdcad2232e7ae21a43ef6","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":76.056,"rawValue":"76.056","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dd65cc71862f092c41376257","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.282,"rawValue":"80.282","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-257282bed877178ed5d11d4d","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":67.692,"rawValue":"67.692","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0c17b6fab10133d6d0849779","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":74.286,"rawValue":"74.286","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dfa0b9108366e6f5b5d144ee","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":74.648,"rawValue":"74.648","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ef7628b4fb5c80e903e1ca6e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":81.69,"rawValue":"81.69","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-924e47440462fc5473ed2cef","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":78.873,"rawValue":"78.873","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-dbe400a5e736a6e85884f2a5","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":83.099,"rawValue":"83.099","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-35b926dfd74c3165bcebf8db","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":76.056,"rawValue":"76.056","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-625ac7066eaba0c884beb5e8","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":67.606,"rawValue":"67.606","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6d7a626168627f97b87e4e8c","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":77.465,"rawValue":"77.465","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-29d48ea7e32ea624b3e816e3","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":78.873,"rawValue":"78.873","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9a79686cfd952f8e225c3c5e","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":71.831,"rawValue":"71.831","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d71db6630ae01022b3c0a8ad","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.282,"rawValue":"80.282","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5cce9f8098aea0789f5d1378","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":69.014,"rawValue":"69.014","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fe842cb4481bd9e173b44dba","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":71.831,"rawValue":"71.831","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8db47f5d36a5513499cd2fbc","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.282,"rawValue":"80.282","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7dcefd71030ae1ddef8a32b2","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":78.873,"rawValue":"78.873","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9588aac6aa4461cd6f1f90a5","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":77.465,"rawValue":"77.465","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2f07c8677914d42734e89685","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":69.014,"rawValue":"69.014","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9866eaabc5eb4a5a64393fc3","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-code-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":71.831,"rawValue":"71.831","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=code_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"code_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-12e05c19977580730ba1b416","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":99.333,"rawValue":"99.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-72c39018e95c5307ad29d713","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":99.333,"rawValue":"99.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-eaeaba0eb5ef30468322d639","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":99.333,"rawValue":"99.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d6cd237212763d1964ec43c8","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":92.333,"rawValue":"92.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-145259d09290f097ea270790","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":99.333,"rawValue":"99.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-cd4018546cf89f77a6f947a0","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":99.333,"rawValue":"99.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-885a8881358c52eb076c34e5","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":99.333,"rawValue":"99.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-cf2c9cd2cad42c93e28cf5ef","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":95.5,"rawValue":"95.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-89f3b95951ea2d311384496c","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":89.5,"rawValue":"89.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-26a68b25f7c3cde310c66def","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":97.333,"rawValue":"97.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-257fcd954cece177c8095a8f","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":99.333,"rawValue":"99.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0c0e1bc4194683dc51130bea","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-842e495f9db24b51bab12e84","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-01ed1dbfe1694ad76acafee2","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-37d45cdeb766dcdb2ad6f6dd","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ccc18475a9ae615756a834fb","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":92.0,"rawValue":"92.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1a26ec945019aeef834cb75e","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c8523be0ec2b32fca1c32a5b","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6013f5dc476105e31eb1446c","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":94.0,"rawValue":"94.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d218f1ccb94508564510495e","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":95.333,"rawValue":"95.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ff9c86c33acee7c75cabc307","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":93.667,"rawValue":"93.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-88f9ffe84bd7c545da2a513d","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":99.0,"rawValue":"99.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a24ac167f6b655a101996836","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":95.0,"rawValue":"95.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1cc886b92fc4345a96c441cc","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":98.98,"rawValue":"98.98","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9dd14a3665da167d2a1fa1f7","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":94.388,"rawValue":"94.388","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-058e944f1d1f212e47106be1","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-66a9211249527f45fb4c4d04","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d6639a45411d5fc22f8a4edf","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":96.5,"rawValue":"96.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-95df62c9b337dcc7502f70ab","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-91549a10d3fbc989dd8b6087","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":93.0,"rawValue":"93.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c22c219f21db6713021b0b1b","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1e03573c0dcd744b63f5e3a0","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1a3db9b2dab36b636e7d973d","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":89.333,"rawValue":"89.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d44e37ea2618d68ac3819225","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9c64e66fe410ee9e4f39a6fd","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9439e8d1e6fa9f7a796d702f","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":96.333,"rawValue":"96.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e4fa360fecbd575714ac1a76","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":77.167,"rawValue":"77.167","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-64056109e47de044124a9d44","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":90.5,"rawValue":"90.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-73de93c6e0b5126b7b45ddd3","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":96.5,"rawValue":"96.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7149a9cf38042cde1c778e19","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":90.5,"rawValue":"90.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a60cb5b6202db1fba820a893","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":90.167,"rawValue":"90.167","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2ccfc870184f250df397cb3b","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-connections","benchmarkVersion":"2026_06_25","metricId":"score","value":94.5,"rawValue":"94.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=connections","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"connections"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5833ef05a18e15944270240b","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":91.36,"rawValue":"91.36","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e0a5146d115fdc2aad5fbe36","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":79.363,"rawValue":"79.363","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d3bfd258d056ea9ddd834508","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":63.467,"rawValue":"63.467","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0317fd1fbfb5eace9a30e147","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":89.521,"rawValue":"89.521","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b0515e6b35fabe54f7873502","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":52.905,"rawValue":"52.905","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-542b49e5969a00e75d4c4bfc","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":77.571,"rawValue":"77.571","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-72db7005c5affdaeefce2c48","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":92.683,"rawValue":"92.683","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ffd0c5cb0077203341a1b25e","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":74.741,"rawValue":"74.741","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b0f61b086829d0041e1b440e","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":59.915,"rawValue":"59.915","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0c61aa261f571b8edc4e99be","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":89.444,"rawValue":"89.444","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-68bc6669e2a22377e0dd94fc","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":90.045,"rawValue":"90.045","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-eac16ef38c363575a3cde7d7","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":75.421,"rawValue":"75.421","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5f8b63b85a27a78ce42afa62","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":90.229,"rawValue":"90.229","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3a1b676ef389c6af2852d03b","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":85.239,"rawValue":"85.239","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6ab75c57e31dcfa4a6856501","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":47.994,"rawValue":"47.994","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6aa2d8e1764a0032aa7634ee","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":21.095,"rawValue":"21.095","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ee8131625035cc1e73ef3911","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":43.824,"rawValue":"43.824","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-be2fd635e49295ab662a32e0","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":60.373,"rawValue":"60.373","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d7f0664ff691c4ad1146cc62","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":79.276,"rawValue":"79.276","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4a4f5db8332ee41d3bca2e46","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":72.309,"rawValue":"72.309","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ffcafaae384a89f4d013171a","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":84.298,"rawValue":"84.298","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f5120bab9e78d473cc66c986","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":88.913,"rawValue":"88.913","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-80ed74102aca8c57a25dca8f","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":89.016,"rawValue":"89.016","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7f95f1a83d76cfc425ee953c","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":62.972,"rawValue":"62.972","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-84069dbbe777219de7bdc251","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":54.301,"rawValue":"54.301","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5dd77a964709446ad5af0978","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":86.19,"rawValue":"86.19","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4f889092847828d94e8518b1","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":88.823,"rawValue":"88.823","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-50a3fe6734232834f95b49c9","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":86.504,"rawValue":"86.504","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-5bf292f1b37c378dab8ff4e6","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":90.214,"rawValue":"90.214","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d5e9aab5a98f3c3bc2f91e5a","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":87.67,"rawValue":"87.67","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d8089548a100d83fed57307a","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":75.536,"rawValue":"75.536","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4b46034c2a9dbb559b6a952a","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":78.148,"rawValue":"78.148","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f9dbe2c9a72e2a352a3812ea","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":51.268,"rawValue":"51.268","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a82ff3b50689cc92e50dee12","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":48.277,"rawValue":"48.277","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fd33d776f6ae13b7480daf13","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":89.758,"rawValue":"89.758","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-022c024a2bb5bc0e36d72345","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":84.614,"rawValue":"84.614","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7e26dab0446558e5045f1471","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":70.395,"rawValue":"70.395","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-10a6326537fa99e2609aecee","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":67.042,"rawValue":"67.042","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b1b7745b5248308c22f46a49","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":71.8,"rawValue":"71.8","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6fae4c85779870e7a5091807","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":84.929,"rawValue":"84.929","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8a0bdaad3e85e8bbf77f640c","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":75.623,"rawValue":"75.623","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ca5ee63b05f2eef20eaff30e","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-consecutive-events","benchmarkVersion":"2026_06_25","metricId":"score","value":87.142,"rawValue":"87.142","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=consecutive_events","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"consecutive_events"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-63002af362b190a651626940","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a89e078689c4a802eaba7b5c","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":78.0,"rawValue":"78.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e559b7acb83025de6ae4b986","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":73.0,"rawValue":"73.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-74bce5e449cc3fe94a88b1c7","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e39cbbacf89c9528fe483bce","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":89.0,"rawValue":"89.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-80cde5b324687e68074f63b4","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":97.0,"rawValue":"97.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e3b142cd623147cb310dd3ca","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":90.0,"rawValue":"90.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-cf224edbf8688ce57e0817d6","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":88.0,"rawValue":"88.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9084a7d3694216f78a7bd0e0","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":37.0,"rawValue":"37.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1ee7f50930a7af617aef703b","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":65.0,"rawValue":"65.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-e2c0afefb5384cef65c3460f","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":76.0,"rawValue":"76.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f12fba682a67fc748496670f","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":78.0,"rawValue":"78.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-46004f126243dfc5998e7a31","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":94.0,"rawValue":"94.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a916aea6ab17e8e957736943","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":78.0,"rawValue":"78.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1568ab941efcc735926c7e6d","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":68.0,"rawValue":"68.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-161ea4d55e1b5a1571325ac2","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":26.0,"rawValue":"26.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d509062abf1201a1cf62178d","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":63.0,"rawValue":"63.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-13753adedfad546b19941fa1","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":88.0,"rawValue":"88.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d4ef3779a7b65cc21d6b0a6d","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":76.0,"rawValue":"76.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1f1dfde088b5028a5d2d42ac","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":69.0,"rawValue":"69.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-436b53fca382bcb9ac36df21","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":45.0,"rawValue":"45.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6894d66809c12f78fc594d8a","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":92.0,"rawValue":"92.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5d466283759b92c47be42109","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":75.0,"rawValue":"75.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-114341c68f175827b2f186dd","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":37.0,"rawValue":"37.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-000370e60e31e0b5f509d119","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":82.828,"rawValue":"82.828","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bbca2dae93f84b122f531fe6","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":93.0,"rawValue":"93.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-cf112448793d1b68a1458274","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":99.0,"rawValue":"99.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bfdf0ca84c67d0ab89edcadd","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-aae11dda322ca1c651066a1b","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4b5e9f6d1e3c238bd1831e9b","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":95.0,"rawValue":"95.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e8c9f7a6eb09ac9e0432ecf4","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":78.0,"rawValue":"78.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-02f755e5a073c70fd03d4e77","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":85.0,"rawValue":"85.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c16678c457c862d69b4d1964","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":54.0,"rawValue":"54.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1ff9416c345af48072d0946b","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":37.0,"rawValue":"37.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4bbaf8b70c63f321a3db5a8c","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":54.0,"rawValue":"54.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-080e5c5c42bddfd9bea9ef1e","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":56.0,"rawValue":"56.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e7741fd43852fef8908305fb","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":52.0,"rawValue":"52.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-933c1be599148b342dc2d2ef","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":62.0,"rawValue":"62.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-285b1e84bdef7ea7dd581ec7","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":59.0,"rawValue":"59.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8e7085a33a660a867842fb53","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":65.0,"rawValue":"65.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0d148df2574d0ef8078228d7","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":59.0,"rawValue":"59.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-539fe191d4721784ae42619d","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-integrals-with-game","benchmarkVersion":"2026_06_25","metricId":"score","value":81.0,"rawValue":"81.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=integrals_with_game","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"integrals_with_game"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-74119fbe0b6d567ba0e243a1","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":68.182,"rawValue":"68.182","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e845c8212f5c8653ce3115a7","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":59.091,"rawValue":"59.091","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-85fe471dbb4c4b34ed726f59","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-759be75dff99fa21eed90da1","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c30de0e66e7b16057bd00096","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":68.182,"rawValue":"68.182","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-76687d67311239182658d63b","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":77.273,"rawValue":"77.273","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-958f2b36ab4ba582c0dff63b","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":54.545,"rawValue":"54.545","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-366c3d758d92ce2209657f7e","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":68.182,"rawValue":"68.182","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-91379d501d7f572be161473e","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":54.545,"rawValue":"54.545","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-27318c712dbbf75928824f39","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-6484785309b58575801a9fbe","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-db0758a5c36104ad0cb11f64","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":54.545,"rawValue":"54.545","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a6e405ca114163dfdd818a42","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":68.182,"rawValue":"68.182","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7dbaf6b95af25ee92b7fc230","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":59.091,"rawValue":"59.091","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f73d9be760bce8a65b1d2acf","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8d71a67dc57ee2ed6d0f0abc","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":59.091,"rawValue":"59.091","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cd4c395b6f7f04642fdc6627","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dfda25ef187dc44ae486cdc8","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":68.182,"rawValue":"68.182","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-76997c1507ea321921d630d6","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d38c1016c80ee1a9fcd10b6a","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":72.727,"rawValue":"72.727","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0ac37e37fe25baec6c0465f7","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8114a2f0dbeefd5d9f2f43de","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":59.091,"rawValue":"59.091","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-fa6656aa8c48c4616964f31c","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":68.182,"rawValue":"68.182","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-528c63dccaf5b001f7f064d7","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4fce12fcf92b4f3a00c3f91f","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-802eadd421a313155ab97ae2","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":68.182,"rawValue":"68.182","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3a856e4739137e02b07d607d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d21d2d60b20f1997806caf7e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b7aad5224461ecaf4ef053dc","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0e0a50f7f29baa842678b2cb","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":68.182,"rawValue":"68.182","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d0f9e2f87476e23e74eeab1e","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":72.727,"rawValue":"72.727","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-87b5336ae631aa591a60955f","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":72.727,"rawValue":"72.727","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-52d0ea5a23265f66f37d5698","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":59.091,"rawValue":"59.091","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-55ad80235f7d4ef932d15d1e","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7db48a25938b24903d507c2f","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":68.182,"rawValue":"68.182","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d2e48c7fa64fc410cf151c01","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":63.636,"rawValue":"63.636","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-52d7202fde255427fa5d2cc8","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":54.545,"rawValue":"54.545","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-db1ff2ae66697e192e13b6e7","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":59.091,"rawValue":"59.091","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2e34c3ffe92e5bc888a923e2","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":59.091,"rawValue":"59.091","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-07936b591078df501225b109","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":59.091,"rawValue":"59.091","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a86cf411b509ebbcf28d9048","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":68.182,"rawValue":"68.182","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c55eb0f68f89af9abd753323","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-javascript","benchmarkVersion":"2026_06_25","metricId":"score","value":77.273,"rawValue":"77.273","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=javascript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"javascript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-99a045be3fe60ec1eb8a5458","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":78.0,"rawValue":"78.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-53178750dc748958209d6aa2","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":68.0,"rawValue":"68.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3dfce42db4d16f4643141c98","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.0,"rawValue":"80.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6ce506005f448c56c1c67e44","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d2ff6b8f2d5592e43e2e237c","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":78.0,"rawValue":"78.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-08e8e97efdeb064d4a0ea54c","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":86.0,"rawValue":"86.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-747af1e9868b2ffbd1c5dc5a","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-5ec7a670f99f8aa6d21aa51b","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":86.0,"rawValue":"86.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d97f062f694a9893d7a90a4b","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":64.0,"rawValue":"64.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-bb5fce4c1211b2acb5072c92","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-ff2e96fd244c09ddb77e1860","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":68.0,"rawValue":"68.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e1da478487de230369bc5e8a","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":64.0,"rawValue":"64.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9d97f4f943c2479148925395","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":66.0,"rawValue":"66.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-81cb1c967328fae9bf2d9a8d","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":72.0,"rawValue":"72.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-941f4c72bd022ada14c13562","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":74.0,"rawValue":"74.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-318a65b11067e530892944bd","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":68.0,"rawValue":"68.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-65d65008e590a3bc2d271a02","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":76.0,"rawValue":"76.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-97993fe56cba8efa0b3b1f93","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":74.0,"rawValue":"74.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-487c7f48c6e81420ea3d268a","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.0,"rawValue":"80.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9b6c90b7e6a0b8025ab990e8","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d49f101c2d7892c11a5387c6","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":62.0,"rawValue":"62.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8772fe657adde42f55e78e26","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":74.0,"rawValue":"74.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f84513c255f70fced9d77bcb","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":68.0,"rawValue":"68.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-27bd83077cc1d43d5545d2c0","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":60.0,"rawValue":"60.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-139656b40593fc06dbc51b6f","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":64.0,"rawValue":"64.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-074d222b7fa5822cb07012f9","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":68.0,"rawValue":"68.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1188b35b7832bd26df12581f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":76.0,"rawValue":"76.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-57337f3151cad26f65fab49e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.0,"rawValue":"80.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0279d874ea2a6160cdffb9a8","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":82.0,"rawValue":"82.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-28199ec0502c6ddbcb69302a","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":76.0,"rawValue":"76.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-077970c01d4beb672163158a","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":72.0,"rawValue":"72.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2ac5d6fd0d6ab9d2eb27d022","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":82.0,"rawValue":"82.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-769e0f5da34f372b6b85efc0","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-5f9a98dc2bee8ba4c1b8d54a","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":74.0,"rawValue":"74.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f4ea8a308a8081a846df4af4","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.0,"rawValue":"80.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-bf547ff02c09e22b6bfa7ca2","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":66.0,"rawValue":"66.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ebed4d52cc7537406ef2839a","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":62.0,"rawValue":"62.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3d447cfd1227210396f9aed8","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-40799742423263cbe381a282","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f992ecdcd6ec1bf24029a65b","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":68.0,"rawValue":"68.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8150a47e7e124ef0ba3a05bd","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2331f3c762f47564409e5e3d","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-logic-with-navigation","benchmarkVersion":"2026_06_25","metricId":"score","value":74.0,"rawValue":"74.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=logic_with_navigation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"logic_with_navigation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-03ce52455d0732faad70a492","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-abeb82f6431083c45816613b","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1bc43abf4a0c4423d26984fe","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9ebdc994ba29a1e451440e75","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-90c760d7c80d3e11613a815e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d80526f1a0f02e9382c27344","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":94.118,"rawValue":"94.118","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3872a55d077bd4462719d709","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":94.118,"rawValue":"94.118","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1d369baeb4bbcb5c19bfbbcd","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e7a3bef152259c8a5a5a8b92","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":97.059,"rawValue":"97.059","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-cdf0ef985494fc79264e8663","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-6487f6d1cd39c1e3e353109b","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":97.059,"rawValue":"97.059","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6f2310373616219f579e047e","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-db8411c70000abb077a13f34","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":97.059,"rawValue":"97.059","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d9bb91ff42b86bb33afe61a2","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b4c8f4f551193268cf8e12ee","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-265f5d9081d51cf17cde06ab","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":92.157,"rawValue":"92.157","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-021a0b8539a7d4fcd0385548","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":94.118,"rawValue":"94.118","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5cb2779bf865a5e0b2707a8e","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-189c6fb20cfca8140d8054f9","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1723f21aad9c3dcbb97b242b","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-229da68ef9c1649ed24a44d4","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":93.137,"rawValue":"93.137","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5e264d28e4d78c0a63511e49","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-80c78b3777b07896cafcd6a8","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-09655dac0c85093c2a3b045d","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":94.118,"rawValue":"94.118","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3200fe37c88dd084da0b589b","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-11c1499d3bd93ac50d82df74","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":94.118,"rawValue":"94.118","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9ee73aeec0f783307ac19fe0","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9b1ad40457ef0d30d40ff4f0","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":92.157,"rawValue":"92.157","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-010648d5e691a7df4c7de692","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9be0a4541d4f9652a3e062f9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-acb70937e09be2b80663ec07","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":97.059,"rawValue":"97.059","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a3001a8a5cbbfa3d6846db1e","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":97.059,"rawValue":"97.059","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-157e4fad37d76d542167dff1","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6c4786f2b419296f08cf24ca","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":94.118,"rawValue":"94.118","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-238971b84d4cbfa526139703","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f5e4b5cde6aa5a9d41eb1d97","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":91.176,"rawValue":"91.176","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1b378a919fc8a446c72e2ce6","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":92.157,"rawValue":"92.157","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-73219947fc6ebdb60a56ba90","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":93.137,"rawValue":"93.137","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7059f2c6f7878edb8810af9c","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":97.059,"rawValue":"97.059","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-91d311a1a217de21d6be0831","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":94.118,"rawValue":"94.118","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-193caa2964a5759d53049f34","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":94.118,"rawValue":"94.118","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-987738cc6622893d492a63ef","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-math-comp","benchmarkVersion":"2026_06_25","metricId":"score","value":95.098,"rawValue":"95.098","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=math_comp","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"math_comp"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4fee1a40277404f05fdcad7b","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":92.845,"rawValue":"92.845","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a5a095fa0601af55ba0c4c3d","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":89.458,"rawValue":"89.458","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-165f05483c6c1d537b4baffe","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":92.17,"rawValue":"92.17","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-40de6f7670b15c33141772b3","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.376,"rawValue":"91.376","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2479f1e4aaefdd1da28efa5d","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":92.227,"rawValue":"92.227","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1ce038220f738ad6f57cbd28","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":92.796,"rawValue":"92.796","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-33403945336ec7a1c5b7ddaf","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":87.858,"rawValue":"87.858","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3fbd277127c5f732c256c337","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":90.671,"rawValue":"90.671","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bf581f39ad71081b1909b225","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":86.531,"rawValue":"86.531","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-29540e4320c19b7456f5c3bf","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":89.081,"rawValue":"89.081","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-906443d64ddeee123ef62c60","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":90.168,"rawValue":"90.168","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-13a3f9c0f85af50a1495d534","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":90.624,"rawValue":"90.624","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3aaa812e22f2d84bd03cda1a","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.286,"rawValue":"91.286","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-70aba71a98342f8d09da80dc","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":92.102,"rawValue":"92.102","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6978336a4a53424865c696cb","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.877,"rawValue":"91.877","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c38ed2aac583cc27650bc61a","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":81.801,"rawValue":"81.801","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0b42abbb6c0f0d22dbf06473","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":90.492,"rawValue":"90.492","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-61f6decb39ec38b44fd26965","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.794,"rawValue":"91.794","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-835700392b3ac0c98d4907e3","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":89.047,"rawValue":"89.047","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-327a89f97a32be720b10d275","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":87.514,"rawValue":"87.514","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-bda96915a01dd19e6410608e","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":89.841,"rawValue":"89.841","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8dbcf65cc1d97fcbcfc78f10","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":88.566,"rawValue":"88.566","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9c924566a62069d1c9674da6","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":86.997,"rawValue":"86.997","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-684e417618ebe4170bbf4e9f","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":85.73,"rawValue":"85.73","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f2565e8821f22a3a83ae0ee7","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":87.982,"rawValue":"87.982","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-486d8724bc394389aca0e5ab","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.474,"rawValue":"91.474","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-241f3a13b8c9a813b92b811d","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.351,"rawValue":"91.351","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9c400eb2f9f2a6a469b3dd69","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":88.646,"rawValue":"88.646","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7fa87c57fa2bf2146e47e78b","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.693,"rawValue":"91.693","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2f95494b6d3f1aef54cffa13","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.535,"rawValue":"91.535","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fd88003515ac9ea4abc4937c","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":89.239,"rawValue":"89.239","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-05f6f62602ec07e7028790f1","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.213,"rawValue":"91.213","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a55f36a7bd439375ad48ebf6","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":90.027,"rawValue":"90.027","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4e640800aa002efc63b893d7","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":90.299,"rawValue":"90.299","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b882550c1b261b48ca8719bd","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.649,"rawValue":"91.649","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a9b70ff396313e27534b70d1","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":85.614,"rawValue":"85.614","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9ad5381f1867756531707c04","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":82.317,"rawValue":"82.317","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5c54a1eccab9c940b0e86d99","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":82.762,"rawValue":"82.762","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-33d6d0b66165304565d933f2","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":86.934,"rawValue":"86.934","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6ee2a2e5e0c55ba48f992e61","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":87.733,"rawValue":"87.733","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-86b02ed1295edec9fc934c11","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.166,"rawValue":"91.166","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-731b9eb4f239a07b668b5cdb","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-olympiad","benchmarkVersion":"2026_06_25","metricId":"score","value":91.151,"rawValue":"91.151","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=olympiad","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"olympiad"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5c50db015afbc91466afe557","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":77.65,"rawValue":"77.65","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8e965b628834ed88c6625164","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":65.667,"rawValue":"65.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e45df10f369da06f82329f0b","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":62.817,"rawValue":"62.817","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d61d1bde9d58fa0835c665a7","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":65.767,"rawValue":"65.767","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-72c4c962dc9f111ef185fc74","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":67.867,"rawValue":"67.867","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8f35f5850c1127a3338158cf","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":65.733,"rawValue":"65.733","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6013e4e0294cc138b516c295","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":59.4,"rawValue":"59.4","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d6131b9a63d8d4ceb237923d","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":61.25,"rawValue":"61.25","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fa26bb7be71b26e9672bf03f","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":62.383,"rawValue":"62.383","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-332a85fad9eed8fe6d99a85b","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":61.7,"rawValue":"61.7","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-3ae9615b9e98804251cb7ee0","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":72.833,"rawValue":"72.833","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-eae7cd7acf75af220349c3cc","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":56.217,"rawValue":"56.217","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d505c77198f3e3b43ff7800a","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":68.633,"rawValue":"68.633","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6e4bc1323533e88330a5e30d","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":79.867,"rawValue":"79.867","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8aa0c6e6301f7272002efb9c","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":75.517,"rawValue":"75.517","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-62d1fa0d06453a4fed7191fc","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":65.0,"rawValue":"65.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4f78a8cb59c34dcb7980a3af","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":74.317,"rawValue":"74.317","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-43817a084c9a663b3fd50798","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":78.067,"rawValue":"78.067","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-404d2178f7f02e3240c0bcef","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":62.25,"rawValue":"62.25","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5e9544cc3cf6325bff7b69be","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":64.683,"rawValue":"64.683","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f89c65f2b811409cb54d5c61","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":47.983,"rawValue":"47.983","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f5323499e1f03a04893bbe49","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":64.1,"rawValue":"64.1","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cac11ebaf650051a0bcdaddb","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":67.05,"rawValue":"67.05","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-88d799e5f82f3ef08182d60f","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":58.883,"rawValue":"58.883","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-824a85c75018a098d760415a","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":66.649,"rawValue":"66.649","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c076549562624dd79c231c1c","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":63.433,"rawValue":"63.433","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-af761000c28379f88b4b0430","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":71.733,"rawValue":"71.733","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c9eb720f32bfdad6a43fba81","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":47.817,"rawValue":"47.817","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a9a53c00bcd6534b1baee17b","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":71.517,"rawValue":"71.517","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-63587fba811d83316052aa80","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":57.033,"rawValue":"57.033","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-dfb0cb4ad7d3abd9c58f6844","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":69.217,"rawValue":"69.217","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6b8bbcb5e1aaae146e9a3667","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":71.783,"rawValue":"71.783","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-672ceb7f221aa584ca01ddb5","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":61.633,"rawValue":"61.633","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-976caacf7e54cc9496be24e8","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":48.533,"rawValue":"48.533","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6b3aa1fbb72ca5e6f69e146d","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":74.367,"rawValue":"74.367","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-159666865b621aff942bf9a9","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":55.933,"rawValue":"55.933","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d9ca9253bc60eed1733be817","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":49.433,"rawValue":"49.433","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b90a6c0f35f8bc1b34f33a67","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":60.617,"rawValue":"60.617","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4807dc6b7cf7ec79852d3b5e","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":72.0,"rawValue":"72.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0df89b930d7462469ec8f9f8","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":74.9,"rawValue":"74.9","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d191c0c3dcc74843ee5f01a8","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":79.467,"rawValue":"79.467","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-30c574db5252a9ebb3693dea","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-paraphrase","benchmarkVersion":"2026_06_25","metricId":"score","value":73.317,"rawValue":"73.317","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=paraphrase","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"paraphrase"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5a2538f6702000be830fcf81","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":78.719,"rawValue":"78.719","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8bdcba592d46fdf5a9cb5ef7","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":66.45,"rawValue":"66.45","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8608f9e240171fb783e27c9d","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":66.476,"rawValue":"66.476","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c2b1d214e4926894e17f0175","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":61.409,"rawValue":"61.409","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bc9dd299fcf9cefc76e09d1c","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":61.645,"rawValue":"61.645","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b61cecd9d5eb45324d9fb768","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":74.731,"rawValue":"74.731","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d1d7234a1309b9de5e756af7","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":56.975,"rawValue":"56.975","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0a316831c15b47c74025ff8f","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":53.411,"rawValue":"53.411","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4623b5e6645cb2912358d702","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":46.872,"rawValue":"46.872","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-f78a5723705575eccf751aba","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":58.192,"rawValue":"58.192","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2734ecf8c0816952b137e812","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":57.746,"rawValue":"57.746","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f445ab16e6ec999531d0b47e","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":56.391,"rawValue":"56.391","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-84e3ee5945f8d4e5db95f64e","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":64.222,"rawValue":"64.222","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7847cc0c9ec65a2676d217ad","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":74.128,"rawValue":"74.128","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b100b566a58424dba2e08896","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":67.753,"rawValue":"67.753","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-72bb67aed820629c6fecca72","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":49.461,"rawValue":"49.461","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-091b87e81453f5494832cf4a","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":63.693,"rawValue":"63.693","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-03c6501f8936517a9813cfee","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":66.365,"rawValue":"66.365","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2cbc7f5ce442435a541b04e5","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":60.725,"rawValue":"60.725","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-164d60f197cbd294ad2d8d30","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":60.236,"rawValue":"60.236","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2a15c1f947efdd7490e4829d","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":60.255,"rawValue":"60.255","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-18e46f17f95518a53a33d3f8","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":58.427,"rawValue":"58.427","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4b4eb94386558670d7b1c803","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":56.035,"rawValue":"56.035","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1a0f9171f89d37f8ff9ff0ca","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":43.875,"rawValue":"43.875","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-559cc66f8201553c8129faa5","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":39.133,"rawValue":"39.133","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c19a2b0f6ccf772f1d0c0fc4","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":65.901,"rawValue":"65.901","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ecfa8c770610ba44a21004ff","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":74.089,"rawValue":"74.089","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-de299cc19377b62a9dbd5cb0","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":51.201,"rawValue":"51.201","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-625e16f4c562df72302af553","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":79.051,"rawValue":"79.051","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6cd7b80a9d867e7cb08a0968","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":69.678,"rawValue":"69.678","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-78570f07ec6e774588740382","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":66.386,"rawValue":"66.386","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-54ed4e07bda96ce8292448ee","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":67.091,"rawValue":"67.091","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ce14e3ee060d3e85d28efa60","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":58.091,"rawValue":"58.091","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b33d0630dc27f1815b1b4775","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":55.717,"rawValue":"55.717","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ed091f24da25c6ed2294db4c","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":72.584,"rawValue":"72.584","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-dcb38d2b912edca630ae1232","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":50.174,"rawValue":"50.174","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d33af1359d68b4cde7fe1bac","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":42.746,"rawValue":"42.746","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f4b907d9f61e3933047b5136","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":52.468,"rawValue":"52.468","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0ccd139cbfd76b8f7d5bffba","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":58.717,"rawValue":"58.717","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ef5fa1164989badc7cc38a17","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":56.537,"rawValue":"56.537","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-86def491612e4ea19c6b11e8","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":55.762,"rawValue":"55.762","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-831c8a70750bf1423e020e6a","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-plot-unscrambling","benchmarkVersion":"2026_06_25","metricId":"score","value":58.562,"rawValue":"58.562","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=plot_unscrambling","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"plot_unscrambling"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a3089444d215a42ca6f86a71","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":65.0,"rawValue":"65.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8a7ab09cb6441bc2cdae92e1","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":40.0,"rawValue":"40.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-db0143672e76f3bc4a36d76a","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dc1c7de7a6af293fb57ffc52","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":45.0,"rawValue":"45.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d23671606f18f7533ebe56b2","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4929132c181abc9b652cbeaf","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":75.0,"rawValue":"75.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ee005b465cf7f1aa927c8a24","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f31778468ac56ae44b046e2f","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":60.0,"rawValue":"60.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-29b90786a8745b0059745c4e","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":35.0,"rawValue":"35.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-9c9e531caab6f5bcd80b87bd","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0ebbe5cc20f2c7c29e1c202d","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":75.0,"rawValue":"75.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0335e203a5a0f39b4834a39e","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-33aeb85012b5eaef19d10877","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-43f847bc9e7eb4b9cdc426e9","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-880eff553de74a24e587d785","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-52d6339e92b405bd2df1340a","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-04115294c5ecc4387b6f71ca","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":40.0,"rawValue":"40.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5733379cac6ba860bf6a7f20","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":60.0,"rawValue":"60.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ce764620b8e6f482a7d0c94c","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":55.0,"rawValue":"55.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1958ce92847a91f4337df67e","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":60.0,"rawValue":"60.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6d2f262ccf6126e63ada7d93","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":60.0,"rawValue":"60.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1f0b28b6f118e000a07f99e8","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":55.0,"rawValue":"55.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-85ae224308cad7fa02613c61","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":60.0,"rawValue":"60.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f0f7a8873dfb3a8fed294d45","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":55.0,"rawValue":"55.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-57d1d01cf6767868e923967a","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-761b2c58e112025ea5dee028","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a5fede18e124abcf938b8ba8","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":55.0,"rawValue":"55.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f812a0e00d6d6f31859630dd","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":45.0,"rawValue":"45.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-39a6ec4a1aab9527ba8e1583","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":55.0,"rawValue":"55.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-57e432206fa62b2c0e0145de","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0fe35a89c2a4b75b1d8294fb","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":60.0,"rawValue":"60.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5693f718f0b58e9203a0189a","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":55.0,"rawValue":"55.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f12509f56f720112a2b813a9","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":55.0,"rawValue":"55.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4e606045a1abadd543a861c8","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1355f27c185458bdb5f1fb1f","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":65.0,"rawValue":"65.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8cb760649e219d156337002d","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":35.0,"rawValue":"35.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c4255c3d49883b36e15699d6","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":40.0,"rawValue":"40.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7d1c29922209234be31937c7","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":45.0,"rawValue":"45.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-56481e2e261b269d8827c6c4","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":45.0,"rawValue":"45.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-447635d31db9fa544e58c591","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":75.0,"rawValue":"75.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9de4a223fadf9e09b7587b0e","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8576831d2a7de67fe938815c","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-python","benchmarkVersion":"2026_06_25","metricId":"score","value":60.0,"rawValue":"60.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=python","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"python"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-21f59a9cfc725e551f3a6c9d","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":72.033,"rawValue":"72.033","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-eed2155809411d0cdb135755","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":54.95,"rawValue":"54.95","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7f68648425373ffa8f226dfd","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":59.15,"rawValue":"59.15","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5fc1ce0d3dd432aaca8f0628","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":65.5,"rawValue":"65.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-94b9c909a50825b7ae9d871e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":62.0,"rawValue":"62.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-5d41e580aa99b5d21a444752","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":61.583,"rawValue":"61.583","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fd24f5709ddce1eb5fddd469","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":62.933,"rawValue":"62.933","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d60785de9fa4748a00871e08","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":60.65,"rawValue":"60.65","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-87c82c9f8dd8d3c2ff7e9cee","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":54.05,"rawValue":"54.05","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1da342a2003d78a8908d69f4","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":58.367,"rawValue":"58.367","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-9cec4cb5ef42cdee6207b4bc","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":65.067,"rawValue":"65.067","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b9fa7d012aa0632574e71ec8","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":55.933,"rawValue":"55.933","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5962b0db159f7eee2db3eeb9","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":61.567,"rawValue":"61.567","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-04eb05f59b28abfbc0f757b2","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":71.05,"rawValue":"71.05","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-068cef08a7ed5d8b87cae023","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":69.1,"rawValue":"69.1","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a8496c4278ca99818f697091","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":68.017,"rawValue":"68.017","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7ef045c602af3d0761ddcaba","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":75.7,"rawValue":"75.7","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-110a7e18bff3d116e23e95ba","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":74.617,"rawValue":"74.617","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-899876b5667680c7c9494740","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":55.667,"rawValue":"55.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b10788e1c505ab3f2e52048b","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":65.85,"rawValue":"65.85","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f385df081e0752bd7d8e904d","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":46.95,"rawValue":"46.95","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-36543ae82f0099cf64912c08","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":54.083,"rawValue":"54.083","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4beaed382137b166bed47f7d","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":60.567,"rawValue":"60.567","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-36fa6bb9eed7d7445fa41d8c","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":54.067,"rawValue":"54.067","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b61a2ea8ee76c2646cd44333","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":62.75,"rawValue":"62.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ec17cb24dae37413c068b30b","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7660d29b07e0e8ee5f31bfcf","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":66.317,"rawValue":"66.317","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bfb066a3e8bd972021e9955b","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":58.667,"rawValue":"58.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-50b6c672e9aa831963f5909b","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":70.517,"rawValue":"70.517","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d7724be7d019afd927793ff5","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":61.433,"rawValue":"61.433","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-99c2deaa9a2b37d282853239","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":70.167,"rawValue":"70.167","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0862e84753dc24e24eeaa061","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":69.333,"rawValue":"69.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f91e7afbdd77d8247bfa3ddc","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":61.417,"rawValue":"61.417","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8e79635d31817291c8117544","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":49.75,"rawValue":"49.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8fced8255d853b25d088a10c","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":66.367,"rawValue":"66.367","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-612d1abd08bddb8830830594","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":56.617,"rawValue":"56.617","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8976ceba9c159fd02635217b","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":50.733,"rawValue":"50.733","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-04bc86f50492861560438e2c","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":51.567,"rawValue":"51.567","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-985ae2a4f6a2893f2f5a0170","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":64.083,"rawValue":"64.083","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8d164c5c69ae4199c2a1e8f0","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":66.7,"rawValue":"66.7","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f13f990d461dcc6cb82a984b","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":70.25,"rawValue":"70.25","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4237d25d31fe3e3ffe1e6003","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-simplify","benchmarkVersion":"2026_06_25","metricId":"score","value":67.35,"rawValue":"67.35","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=simplify","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"simplify"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8fd000a6377dace9307aeaa0","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c3efdcf67e75761134c4292d","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6f72222157d69480bbd3ebef","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8ae6884b299cef80a8b93e92","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2f44b92a1390f3f25d09b6af","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1ba01e394873afd2cc163e41","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c4474adef838913a4e67b2d3","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-719bc7a368a4364bfac35da7","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bfadc4f4850a6adaad7482b9","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8b87f97a4682215bcccbfe5a","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":90.0,"rawValue":"90.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-72f264f21fe16c6f30b15e31","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8f0cf2c414211cab0d9f0575","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":94.0,"rawValue":"94.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3e50a6b7823ca87db1a42cdd","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b183e943cbd66d26aa990b4a","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0ae3e1fdb5f9e21ce37465b7","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1e3ddf91d56c57a0470bc4ef","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":90.0,"rawValue":"90.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7e8fd2fc85d6b7efb60c81f9","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ff3787d417f50d06bdfa013f","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d5706c432f159e46ec65329e","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":94.0,"rawValue":"94.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-563ecf217afb3e790c597459","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1a55fa514246a27999de4d7f","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fc1b8a7dd7fc94ec82b4953e","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9928ec256ad633d9643941bf","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":94.0,"rawValue":"94.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cb1e0c7c55b20810e3f4e45f","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":88.0,"rawValue":"88.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-47b3afea9d180ed6596ae6c1","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":82.0,"rawValue":"82.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1abc9d697bfc5d525237b6ce","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a8951823249783f75e29d0d4","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-af890d32500d01714648b0cf","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1a6712c764f977808e8bb7f7","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-775efe9497b09ef914f119c7","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a3a2a903694db9a4cd578d59","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5b7fe67cfcf9e35176d155e9","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c5ab844070c634686aa05199","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":94.0,"rawValue":"94.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-24494b3252ad76af20970f70","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":92.0,"rawValue":"92.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7a86a68255f9200e419a3d74","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-175907b4e1db8c9065aaf750","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":94.0,"rawValue":"94.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6023b073f8551ce5285c52d1","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-89bfe706e82d1e04960219ee","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ae036f5cbfcde33d6c2c05f7","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c88b855e0de8b91ea03d5dfb","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a8ac15535f7b6cf6312ad2c8","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ca9c77e6a059e424aac86741","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-spatial","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=spatial","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"spatial"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-45591e44f49a50353faad1aa","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":71.7,"rawValue":"71.7","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-702c143843acd91da470fa89","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":65.85,"rawValue":"65.85","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3af2829854f825d3b7c0845a","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":66.133,"rawValue":"66.133","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a1c336b7ae254ac07b649b2d","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":69.4,"rawValue":"69.4","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7c79cdb8f804d4d4acd7cf35","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":80.6,"rawValue":"80.6","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3522b2428a0243eaf13c5609","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":61.317,"rawValue":"61.317","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-eab9164bd60c9ae20669f8c1","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":68.9,"rawValue":"68.9","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-cc1dd3cf112a504a3724fb5d","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":62.317,"rawValue":"62.317","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-56b446bc152be6b429a53a11","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":67.267,"rawValue":"67.267","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-b35e868755fd55cb7267c08f","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":70.5,"rawValue":"70.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-461aa23fc10139b4a4b424e2","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":73.833,"rawValue":"73.833","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-96dd6829d9f59293c33457a6","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":67.75,"rawValue":"67.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2bbad1b16bdc23d8af4e3d18","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":74.933,"rawValue":"74.933","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-499f59eff34653e02564b897","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":77.25,"rawValue":"77.25","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ec934995d4ae7fe1e062f0ed","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":75.2,"rawValue":"75.2","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7a427ce5f2f98afe8a13a050","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":67.467,"rawValue":"67.467","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-279861201a52b3fee00f912b","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":75.6,"rawValue":"75.6","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-61385429b88715d2dc57cd9c","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":82.25,"rawValue":"82.25","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1a828791d269123854d9d0df","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":62.883,"rawValue":"62.883","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-21fdd711dbe7c2d2895807e8","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":72.133,"rawValue":"72.133","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-417cc60766ee8b6401972723","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":59.883,"rawValue":"59.883","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-10ee948443019a84e38b8582","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":65.7,"rawValue":"65.7","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-63ad00aca489b0c8f8fa42d4","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":74.783,"rawValue":"74.783","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ef814f8cc9a0f97924968067","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":66.633,"rawValue":"66.633","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-27c981d9e0e7e680e45fa9ce","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":67.653,"rawValue":"67.653","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bf6e07e92e918fc40fef0156","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":79.017,"rawValue":"79.017","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-559ec8bbcdb03b8c2393c7cc","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":72.55,"rawValue":"72.55","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-afdd7b4aec16a5b3a94b6d13","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":69.067,"rawValue":"69.067","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-045a54cb7c2dba2a8b25e0b3","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":75.917,"rawValue":"75.917","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e6131b2bf321adc400b0c454","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":66.833,"rawValue":"66.833","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c247c44464d7f85224bbe4db","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":73.583,"rawValue":"73.583","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-94c0389c9a409fac7978cfd0","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":78.267,"rawValue":"78.267","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-43bbfe103c07b9ee809aee3a","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":66.883,"rawValue":"66.883","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-207bd018a3585a03c05a4a5f","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":68.567,"rawValue":"68.567","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-144faceed55b74ae33c5957b","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":75.367,"rawValue":"75.367","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-94dd8ba4b34b5cebb388c7ed","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":63.55,"rawValue":"63.55","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3f2630060dd6a78c3c004116","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":60.133,"rawValue":"60.133","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-691552cd94d5d8a3caf3a17d","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":56.867,"rawValue":"56.867","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5c88bab6ec189c3d9e4c4121","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":76.283,"rawValue":"76.283","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8cf6a05aa6e0a39a6a513254","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":73.1,"rawValue":"73.1","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8ab7f86b3780bd4b53678f86","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":81.367,"rawValue":"81.367","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b25f3ce30c7033d3efc6648a","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-story-generation","benchmarkVersion":"2026_06_25","metricId":"score","value":76.517,"rawValue":"76.517","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=story_generation","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"story_generation"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e8e5ca2a675775ed9035f5a2","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":81.7,"rawValue":"81.7","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6d1ec17d5d858c0b33b01c41","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":63.717,"rawValue":"63.717","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4ced612d9ec3bbcb8b147797","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":65.15,"rawValue":"65.15","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2e8461aae355f0c7992081ae","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":66.283,"rawValue":"66.283","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-607ffd34242c657c71a7e229","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":77.667,"rawValue":"77.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f1ec528c7721646bd318a322","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":66.433,"rawValue":"66.433","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-5f4bec9673a497e2b8c42369","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":61.65,"rawValue":"61.65","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-dc2285e84269cbc3bcc176c5","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":71.217,"rawValue":"71.217","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a4501e8a38a7eecfc59d9258","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":68.85,"rawValue":"68.85","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-858ad5b323b8994c46e59eaf","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":71.5,"rawValue":"71.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-6dced55154cd239608ca41e9","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":72.1,"rawValue":"72.1","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-abdb37f5515ac616ef6e4d92","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":69.5,"rawValue":"69.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-786178516d61f4981a92fbd8","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":65.667,"rawValue":"65.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-302bbfdf6ed7fec358730e31","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":88.233,"rawValue":"88.233","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1d0f7606ae79177455e23a77","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":82.583,"rawValue":"82.583","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8c0476f1a69e592776e4b42f","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":68.467,"rawValue":"68.467","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1dfe5e20d4f89e9522e62d76","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":75.85,"rawValue":"75.85","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-825eab9e5721966c812af04b","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":84.767,"rawValue":"84.767","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-aa63f60ab418f3f8fa0e9e32","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":68.367,"rawValue":"68.367","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5bf09074409a6965226b3fe1","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":74.55,"rawValue":"74.55","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-580af7e188d2ab6979396655","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":56.467,"rawValue":"56.467","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7ea92aea2cc9f98bc4ea1efa","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":63.2,"rawValue":"63.2","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7622d4e73c47b297d60d578d","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":63.383,"rawValue":"63.383","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b526081e0c25bd7e76109e4e","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":59.633,"rawValue":"59.633","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6c28880daf6b19b5494e0cf8","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":71.767,"rawValue":"71.767","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b135c4f6cf429e8450dc1751","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":68.417,"rawValue":"68.417","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a738a99a22d5cdb1ccb26ca7","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":72.317,"rawValue":"72.317","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-5f78ef5babc08023f1514f5a","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":64.933,"rawValue":"64.933","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3b03360e41de70b5d6f3a1f1","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":69.433,"rawValue":"69.433","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-5c5cd36e180200f5c4a6d9a9","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":73.167,"rawValue":"73.167","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-905aa659c10e4a0e96d32587","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":73.15,"rawValue":"73.15","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1190b9252bd7d319f290a0ba","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":68.083,"rawValue":"68.083","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a5bd101426a746d8f5db197e","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":67.5,"rawValue":"67.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-21daab08a8310a71bbf4c50c","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":58.317,"rawValue":"58.317","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-edbf2d5d45fd15619ec6ff61","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":69.35,"rawValue":"69.35","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fba6858ab982e0c8a8084f4c","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":53.933,"rawValue":"53.933","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7dbb7d2fdda54e64fd746528","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":52.617,"rawValue":"52.617","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-857f6468ad73732dcb8d8aa3","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":64.317,"rawValue":"64.317","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8bb96623da5c0b5382465be0","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":83.8,"rawValue":"83.8","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-567e2c8a7397327d2c098650","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":75.933,"rawValue":"75.933","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e3a76a7a07d0001c4010c4b4","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":77.35,"rawValue":"77.35","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-35f5b20384ef548b37a12cec","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-summarize","benchmarkVersion":"2026_06_25","metricId":"score","value":79.15,"rawValue":"79.15","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=summarize","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"summarize"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3dcc80b7689e250ec79fccc0","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":56.135,"rawValue":"56.135","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1481dbfbb1dec8e58c82875a","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":45.923,"rawValue":"45.923","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-36f9c9835edec73759ecfd2a","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":48.173,"rawValue":"48.173","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1e02ab8350e47f8a53f8f550","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":47.231,"rawValue":"47.231","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-17487fcdd86fbe2506ef3e8f","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":51.077,"rawValue":"51.077","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7e88576fe8ced687ea87af84","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":51.962,"rawValue":"51.962","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9db6b5caf4324b9374cd22e6","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":45.077,"rawValue":"45.077","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-045690153ce3f2e7b41c0bff","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":42.442,"rawValue":"42.442","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-27f4874fcb5cf6ec562712e8","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":46.115,"rawValue":"46.115","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-a85c4f6f876bc5fadd47d7fc","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":48.538,"rawValue":"48.538","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4cdb51f4f289e6fb4c38b7f3","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":50.346,"rawValue":"50.346","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f5e418470acf15ed1bf9b0a4","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":48.192,"rawValue":"48.192","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-695ed9feb0219aa3f7be9404","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":49.462,"rawValue":"49.462","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-80938f578993741e90d4b0ec","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":52.346,"rawValue":"52.346","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f788e67f1fca02d3292ee8cc","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":50.5,"rawValue":"50.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-02651e5fd834c9cb619d7aa9","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":40.615,"rawValue":"40.615","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-17dee8829c1d8df6ed97c629","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":47.135,"rawValue":"47.135","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8787ac97c5108a46c5d4d616","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":45.481,"rawValue":"45.481","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b8ccf63f6a8a80bb963d2e75","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":45.865,"rawValue":"45.865","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-9ff9ad8dd2350bd8ff676831","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":42.346,"rawValue":"42.346","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-71903b464be0c577c46b6075","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":48.827,"rawValue":"48.827","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-69d1ec10ae03519fe271c548","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":45.577,"rawValue":"45.577","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-80f2652b6208d6203539fab0","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":45.596,"rawValue":"45.596","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4289c1ec6a3486e205d93ed0","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":49.385,"rawValue":"49.385","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-aab6f1b077c360dd077029c7","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":50.588,"rawValue":"50.588","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f100bcf8e1d8b7d9ec71379a","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":51.75,"rawValue":"51.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ce4022eae9ed5c0e8ec3630a","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":55.904,"rawValue":"55.904","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ab60e300f64b5f3a08d87a9d","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":47.596,"rawValue":"47.596","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-dd5221929a54b2bec155dd4e","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":49.308,"rawValue":"49.308","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c57a7ed49b4c0be5e94ba413","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":50.25,"rawValue":"50.25","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3d53d85d996ac88391ea2f64","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":43.577,"rawValue":"43.577","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bd5f028a06827ca1e7fcfa94","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":43.423,"rawValue":"43.423","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-432757202a6e46acb53a21ad","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":46.096,"rawValue":"46.096","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7d5f525bb575f2cbd797bae9","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":47.538,"rawValue":"47.538","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-15cbcf4ad4313775499bc1c6","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":48.404,"rawValue":"48.404","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-54071ed178f4314df860705e","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":45.846,"rawValue":"45.846","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-bf57a379c07d235881b65b73","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":42.846,"rawValue":"42.846","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bfc53457b8a58ab19df64a23","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":44.654,"rawValue":"44.654","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-43e9e74a9a0437a0896c55c3","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":45.519,"rawValue":"45.519","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-92cc580865c38d195fc3fa7d","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":48.75,"rawValue":"48.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-030725f4aee61ad3520aba10","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":51.019,"rawValue":"51.019","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1e0e16be541530b68545a86d","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-tablejoin","benchmarkVersion":"2026_06_25","metricId":"score","value":50.058,"rawValue":"50.058","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=tablejoin","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablejoin"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5d686df9a5a9e828b622d2a6","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":94.118,"rawValue":"94.118","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-91f97ae94f2e63ccd7789dd1","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b9c5b72858f1745ca1593deb","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9747a7d33757016e0d508bfd","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-917bcfd18751b92c775ae1af","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":94.118,"rawValue":"94.118","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-405c28d811b05874cb873e39","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":94.118,"rawValue":"94.118","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-373b1f5c44253f92bcbc0887","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-180baa7ba22a07f652b26255","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-539521f6a9229abed2475afe","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2250e2f4309199e417530f19","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-48a8eb4efd54fa4f29039fbd","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-647334cc7eb4223b879bf9a5","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3e4d59278a5df7d1b3b66e34","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8c371173111cd251dbff742a","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2c92cf765d6e8b161758c553","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dd34def05d8cbd69cc0da29f","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-72b75648358753856ce5fa9e","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6c0aaf46b385029e530f38ba","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-88495e452dabecbd2c1ea083","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-68320be2c4463d8ff862bd38","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f6350824bac8e9f2a68e2843","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a802a3be0bec145ce169506d","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f70982ed1b0c05a15fee1a67","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f924be6e15ca66b36360d371","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-70e9d99d3acd42f6ba8b2f8a","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5b0870b059206f45fa417848","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-708706e7811da6408cc85f79","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-144c036c92e8e85a513fc348","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a7cb4a7360e67b5798434a20","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9f2dcbef41765b864c6e107c","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8d0b0ff165fa6a13d7ecb5db","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-03fffd31dff49eefc9edd86e","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f8eca8a3ba8156504a5a97f4","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-796ddacf855f66be7c523bd6","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":92.157,"rawValue":"92.157","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f1ef46d018f172c60f752c1c","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-07fe3c5e228af828f38f68fc","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3ee416bcc2b78119d0207649","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d85cda69cd3d559e977d6c76","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-01808d297c74a60d57790bb5","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-27bf4b8029be9f98876084bd","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8ffe1a0f6d863e65ab5925cb","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":96.078,"rawValue":"96.078","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-79b95672e255debbda32cd58","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-tablereformat","benchmarkVersion":"2026_06_25","metricId":"score","value":98.039,"rawValue":"98.039","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=tablereformat","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"tablereformat"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-12f49ce0f8821b18ee3257c6","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":84.615,"rawValue":"84.615","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-28bebbc687c0894248a134f1","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":78.846,"rawValue":"78.846","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cdb7f24f9c54666dd295a36b","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":82.692,"rawValue":"82.692","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ee6149d0802449eb2ede2e5b","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":80.769,"rawValue":"80.769","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-73656c46fa925e494a7b6183","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":80.769,"rawValue":"80.769","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-865ea429f2d3d0eb7f158169","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":78.846,"rawValue":"78.846","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7245b10588d76863511c8f26","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":73.077,"rawValue":"73.077","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3af822d5cbc9d359160bed8a","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":80.769,"rawValue":"80.769","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b87931be996ee1c1f652cc7a","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":73.077,"rawValue":"73.077","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1559be248c4e16da0d4f9b4a","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":86.538,"rawValue":"86.538","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-2801a29a8ad68911c5bcf350","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":78.846,"rawValue":"78.846","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-445440de9911fe31201a5a49","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":80.769,"rawValue":"80.769","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6ff76ce20cfcd321298ba3c3","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":84.615,"rawValue":"84.615","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4b2eee03adabdd21e05f0148","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":80.769,"rawValue":"80.769","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9e56e6bdab7814ba6666b86a","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":80.769,"rawValue":"80.769","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-855301fb50d157d8d836c471","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c8f2d2337acc7b852326c273","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":78.846,"rawValue":"78.846","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d64f78abab51bf850381e38d","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":82.692,"rawValue":"82.692","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ad689680aa6bdd09279419bf","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":75.0,"rawValue":"75.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-93522971f29e776d8b78ef9e","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":88.462,"rawValue":"88.462","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-5e72a3a063cc8b78c0338913","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":73.077,"rawValue":"73.077","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ec61b2866e24e12b867e46be","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":78.846,"rawValue":"78.846","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-72a55783968d34a41d14542c","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":78.846,"rawValue":"78.846","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a50170a7a9b917c8c95c70ed","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":76.923,"rawValue":"76.923","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1ae832b349e1787452255e97","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":80.769,"rawValue":"80.769","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a705eaa8a9ad0ad2e64aa1d7","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":88.462,"rawValue":"88.462","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-593438d0c382ebd0112002fd","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":84.615,"rawValue":"84.615","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-296d0dbbc1ff7cb8f9314ab0","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":73.077,"rawValue":"73.077","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-04fb3124c4ccc6732da8ca75","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":84.615,"rawValue":"84.615","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-5c22bb0ed1c3260e10091d35","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":86.538,"rawValue":"86.538","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e6df08b1b7a11fc40fd2b8c4","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":82.692,"rawValue":"82.692","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-816a144d6b5cc472aa767256","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":86.538,"rawValue":"86.538","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-19c4e7805ceca3b99435999a","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":75.0,"rawValue":"75.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ff9e912b0f46df7be4a5ae0d","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":69.231,"rawValue":"69.231","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c9fc3ffdb22cea2010c6965f","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":82.692,"rawValue":"82.692","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6d39cdcca601c68257c54772","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":76.923,"rawValue":"76.923","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-cc0465122d95b4e1354d45f9","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":65.385,"rawValue":"65.385","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-54f3edde13df683a98c52c0a","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":67.308,"rawValue":"67.308","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d8f5998c0c0882088773d392","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":78.846,"rawValue":"78.846","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-daab61522fe074a70725096f","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":59.615,"rawValue":"59.615","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ab6104df2a3042fd04542761","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":80.769,"rawValue":"80.769","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-31dad745c7a5d044d7026745","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-theory-of-mind","benchmarkVersion":"2026_06_25","metricId":"score","value":78.846,"rawValue":"78.846","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=theory_of_mind","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"theory_of_mind"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fc02427935d08c13e2389924","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":53.333,"rawValue":"53.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-67a2855da4a5607dbd69b623","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":20.0,"rawValue":"20.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-be5d7b27bb4b3b033e809eb8","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":33.333,"rawValue":"33.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-08f8c9a73eff2ddbe9c16c3a","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":43.333,"rawValue":"43.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fa9e2d144b0661b8d88798de","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":33.333,"rawValue":"33.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e1c9a24ddc895cbf9e863f8c","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":43.333,"rawValue":"43.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ed8ea7c38242c23fea6f1a41","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":23.333,"rawValue":"23.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-79a674e10f865659bbef67be","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0bf06bde22c000d4a273654a","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":23.333,"rawValue":"23.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-7cae398d0292d08a6255babd","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":26.667,"rawValue":"26.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b56f9625d448e7ab10110812","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":56.667,"rawValue":"56.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0734d4004b86609256bf0941","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":23.333,"rawValue":"23.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ae6ec69f758f692aa6e06fea","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":46.667,"rawValue":"46.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-329aa6e9c34e18049be50b04","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":23.333,"rawValue":"23.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2a994f489b884b2a7c6d1921","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":33.333,"rawValue":"33.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7e6d415a0dead0f4b0b22b40","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":26.667,"rawValue":"26.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-08fdf9f677d6f155ac11a80d","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":26.667,"rawValue":"26.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-89937b8494ac3506146e59a1","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":46.667,"rawValue":"46.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e57af9a0379e696caf463ab3","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":36.667,"rawValue":"36.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-16e8192b4d460807a2df7c19","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-db1d2855cc51cdf6a7f715d5","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":46.667,"rawValue":"46.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4c986b5270ad1e20785b4faf","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":36.667,"rawValue":"36.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8a42a17bf4e2a680b6557c83","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":20.0,"rawValue":"20.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-79b1878a978b88249bbec360","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":20.0,"rawValue":"20.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-234ce01e9dcb23191bc3f157","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":26.667,"rawValue":"26.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-afac7d38b9dd5c5c6d9f6fab","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":43.333,"rawValue":"43.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f3daed7b5bcf25136aa3744e","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":43.333,"rawValue":"43.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-56ef582db0c627c6a8a2611e","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":36.667,"rawValue":"36.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-0b9d137e095629da92b6b7ce","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b670c095e6b63e2f7ff2f139","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":46.667,"rawValue":"46.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a11b1ad9350cf20fe7c9e5d3","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":36.667,"rawValue":"36.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-763adcad8c7248c25fd41a3f","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":43.333,"rawValue":"43.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-69213a2395cb791da96da84a","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":26.667,"rawValue":"26.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b52b409391e51eb2399bbc5b","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":23.333,"rawValue":"23.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4a700006a72b3bfd113b15a8","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":53.333,"rawValue":"53.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1d44fed410044847a50cf3cd","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":23.333,"rawValue":"23.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-54b6be2e3df15b1da5a882f3","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":23.333,"rawValue":"23.333","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-426a57197c534ac619824915","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":20.0,"rawValue":"20.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dcc50f42a94e3ffcd1ab4044","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":26.667,"rawValue":"26.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-835221793fb2e9bc4e4415ea","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":50.0,"rawValue":"50.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-47d3f0c1fad0588a3a13c069","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":46.667,"rawValue":"46.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-cfe4fae90143c7aa1177994e","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-typescript","benchmarkVersion":"2026_06_25","metricId":"score","value":56.667,"rawValue":"56.667","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=typescript","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typescript"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d8f53fc811e93c6987ea382c","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":94.0,"rawValue":"94.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d50485487a6a6dcf4b15a794","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":78.0,"rawValue":"78.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-47c87d926e7895f420510e8e","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c70c318153fa6c3abf4bef32","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":80.0,"rawValue":"80.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2738de63197644e0bb68994f","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":78.0,"rawValue":"78.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-15b58897be3e47474583dc47","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":92.0,"rawValue":"92.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-bce0d49d4c65ca59cc7d65c2","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":72.0,"rawValue":"72.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1a674a6b6030c4aafd8ca934","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":76.0,"rawValue":"76.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b9abc6b994abd8a9b4c79201","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":74.0,"rawValue":"74.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b41a95ab2a7f17e13c23e25a","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":82.0,"rawValue":"82.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-52c3f09af03a649205924d92","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-791effc36acce69762e7dd7b","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":80.0,"rawValue":"80.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-a460f0232f7ab55928cc5b3a","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":82.0,"rawValue":"82.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7aa926224dabc164f78f3e18","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":82.0,"rawValue":"82.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-899d6b771191bf2a6f69fa18","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":86.0,"rawValue":"86.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-04bf2d6a075b075d80faed2e","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":74.0,"rawValue":"74.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ee322baf2c8329251839a692","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":88.0,"rawValue":"88.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d05134c19aab0e640b373acf","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":90.0,"rawValue":"90.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5a345155f80d24a5854f9027","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":74.0,"rawValue":"74.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1a1f1c771f9bed3ed634cfd5","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-2d409214e2ddeaf2ce171539","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":78.0,"rawValue":"78.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6cb8e08cea50c71716586d20","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":82.0,"rawValue":"82.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-aafbfc140104e72df84ca74f","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5b1c8cbd61a6f9661ad3e851","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6a6fdcd369cf0c9671f8e209","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":54.0,"rawValue":"54.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5cd11067c87eb2e9837d6582","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":82.0,"rawValue":"82.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c6c611ad7e9267155c7c691f","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":88.0,"rawValue":"88.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4cb45af04faee8628144f069","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-e22ffd47e073ac5f97f6ab71","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ed91b43ea3b9b07a4fb6e2be","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":86.0,"rawValue":"86.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a1faa7bf0d920cc2c11c1b02","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-85de14795e47ac3a5aaad168","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d073ec90334adbb6d95a1d5d","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":78.0,"rawValue":"78.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b00d1d8369af19b9a0b3a813","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":80.0,"rawValue":"80.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8857681292a45cb3d906acbe","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b03f77954f9964529fc73d22","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-de5c0a1ea18b15deb823204e","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bd1724de55fc47c081f800d3","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":82.0,"rawValue":"82.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4e83f7ba8a205171173f401b","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1d832f2cc6e531da8a8aecb7","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":76.0,"rawValue":"76.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-1944a34e380412f6f5595a8d","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":78.0,"rawValue":"78.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8f635492eb630a8fe710bc46","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-typos","benchmarkVersion":"2026_06_25","metricId":"score","value":86.0,"rawValue":"86.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=typos","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"typos"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-841965cd64777394b35b567b","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5-max-effort","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=28;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-8bd85b5cfd4e7628b102b405","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101-thinking-64k-high-effort","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":77.5,"rawValue":"77.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=0;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2a70e84b009c037c69abaff9","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6-thinking-auto-high-effort","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":94.0,"rawValue":"94.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=1;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-dbf70e13d7419c3c38318bd9","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7-xhigh-effort","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=2;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3f9c942f3bc684c257d2de5e","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8-max-effort","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=35;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-fd93c8de6e4f1e9e56fdaf8b","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5-max-effort","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=34;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-68f3a7a000a226b3e4448b8f","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6-thinking-auto-medium-effort","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=3;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3135999f25fcff2a93ad667b","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5-xhigh-effort","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":88.0,"rawValue":"88.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=4;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-33f94d65e0c30274ac386206","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":49.25,"rawValue":"49.25","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=5;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-17d06b9e90ef5d2afd6b165c","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=36;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-54627c6891c29bca8fc1903d","canonicalModelId":"deepseek/v4-flash-vision-exp@current","modelName":"DeepSeek V4 Flash Vision Exp","modelRef":"deepseek-v4-flash-vision-exp","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":98.75,"rawValue":"98.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=45;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8ebd10a388f0065f5bdfe904","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":92.0,"rawValue":"92.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=6;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d213716b45763cb8e9831f0a","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":96.75,"rawValue":"96.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=40;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-33bab6bce4abd21ba5cb2b9a","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview-high","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":85.25,"rawValue":"85.25","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=7;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f7dba1b4b412213bf11c28d8","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash-high","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":77.25,"rawValue":"77.25","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=8;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ce9324cfe01e37c93097434e","canonicalModelId":"google/gemini-3.5-flash-lite@current","modelName":"Gemini 3.5 Flash-Lite","modelRef":"gemini-3.5-flash-lite-high","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":32.75,"rawValue":"32.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=32;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-eaa77d100747ec8b25d52327","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash-high","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":85.75,"rawValue":"85.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=33;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2d662589e16219cbc26d4189","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash-high","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":94.5,"rawValue":"94.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=42;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-56d6d2d8a5da76bf3c8ae7d3","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":65.5,"rawValue":"65.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=9;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-17aa327266136a7ff121d680","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":88.75,"rawValue":"88.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=46;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f3822ebbbea2988c2fb8107d","canonicalModelId":"zai/glm-5.3-flash@2026-08-27","modelName":"GLM-5.3 Flash","modelRef":"glm-5.3-flash","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":79.5,"rawValue":"79.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=47;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e6b81542db577c8d0d18c155","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11-high","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":84.0,"rawValue":"84.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=10;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-cb1fff47d9347127c1ffaa87","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"gpt-5.2-codex","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":70.0,"rawValue":"70.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=11;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9251b93dcafbfd59bcff7265","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-xhigh","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":60.372,"rawValue":"60.372","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=12;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-f277b80a63a9525a08b398c1","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-xhigh","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":97.619,"rawValue":"97.619","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=13;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2bb6b39e731891ee1fd6a455","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-xhigh","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":98.0,"rawValue":"98.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=14;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3a1e059bb4fbdbb12b585254","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-xhigh","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=15;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-593668a075c6b909959a3e8b","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna-max","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":93.5,"rawValue":"93.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=27;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-878e675a42f00d5dd1fbb888","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol-max","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=25;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a31b098197e09e3f1093feb0","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra-max","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=26;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c3b3d2e0885f3ad52608f8f7","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":94.0,"rawValue":"94.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=17;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6833f9f2f08b162398dbc02a","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":95.5,"rawValue":"95.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=41;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-945c4c69a7ce6eeb1e8107c1","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6-thinking","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":78.5,"rawValue":"78.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=19;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["heuristic_model_mapping","source_version_label_without_id"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6e9771376ec70307a7a657d8","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":96.0,"rawValue":"96.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=20;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-530eaa0d53ca460a76ee6c59","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=30;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-47df159992b363c1a154a2d8","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"minimax-m3","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":61.0,"rawValue":"61.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=21;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-938e833e1abc169734c5b815","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"qwen3.6-27b","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":53.75,"rawValue":"53.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=22;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8514f0d2c525457b23ef7623","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":68.0,"rawValue":"68.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=23;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0be0c43c0040c41aca60e8f7","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":74.5,"rawValue":"74.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=24;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["curated_model_alias","source_version_label_without_id"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ec56896f17d6b5814273fe75","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"qwen3.8-27b","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":96.5,"rawValue":"96.5","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=43;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-51eaed25b9abc8279b2d8e5e","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"qwen3.8-flash-next","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":98.75,"rawValue":"98.75","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=48;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-161a2f8e6e6cf7a52617f012","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max","benchmarkId":"livebench-zebra-puzzle","benchmarkVersion":"2026_06_25","metricId":"score","value":100.0,"rawValue":"100.0","unit":"percent","evidenceLevel":"A","sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourcePageUrl":"https://github.com/LiveBench/new-livebench","sourceApiUrl":"https://api.github.com/repos/LiveBench/new-livebench/git/trees/main?recursive=1","sourceLocator":"path=public/table_2026_06_25.csv;row=37;column=zebra_puzzle","retrievedAt":"2026-08-27T06:23:04Z","observedAt":"2026_06_25","payloadSha256":"d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","protocol":{"harness":"livebench-official-table","release_date":"2026_06_25","task":"zebra_puzzle"},"harnessId":"livebench-official-table","qualityFlags":["source_version_label_without_id"],"subjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d18e3d06784c2ced3b8ee1d2","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"Claude-Opus-4.6","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"all","value":63.11,"rawValue":"63.11 ± 0.44","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=11;column=All (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2026-03-06","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"AIBuildAI","split":"all","subject_type":"system"},"harnessId":"AIBuildAI","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1b7897cc32e211442b53e573","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"Claude-Opus-4.6","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"high","value":46.67,"rawValue":"46.67 ± 0.00","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=11;column=High (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2026-03-06","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"AIBuildAI","split":"high","subject_type":"system"},"harnessId":"AIBuildAI","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-922f012b1b271363e6c68bdf","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"Claude-Opus-4.6","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"lite","value":77.27,"rawValue":"77.27 ± 0.00","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=11;column=Low == Lite (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2026-03-06","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"AIBuildAI","split":"lite","subject_type":"system"},"harnessId":"AIBuildAI","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e8888442506efed604e62065","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"Claude-Opus-4.6","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"medium","value":61.4,"rawValue":"61.40 ± 0.88","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=11;column=Medium (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2026-03-06","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"AIBuildAI","split":"medium","subject_type":"system"},"harnessId":"AIBuildAI","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f7e2e5cfc89d3f4ab5b6d38f","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"all","value":43.56,"rawValue":"43.56 ± 0.89","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=22;column=All (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-10-10","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Famou-Agent","split":"all","subject_type":"system"},"harnessId":"Famou-Agent","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d844f54714fc905979ad4f3c","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"all","value":59.56,"rawValue":"59.56 ± 0.89","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=15;column=All (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-12-27","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Famou-Agent 2.0","split":"all","subject_type":"system"},"harnessId":"Famou-Agent 2.0","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-8bb1a3c8386b53974ab612a5","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"all","value":38.67,"rawValue":"38.67 ± 0.77","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=24;column=All (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-11-03","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"CAIR MLE-STAR-Pro-1.0","split":"all","subject_type":"system"},"harnessId":"CAIR MLE-STAR-Pro-1.0","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-125b7b9b6a8ca1a845b45925","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"all","value":44.0,"rawValue":"44.00 ± 1.33","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=21;column=All (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-11-25","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"CAIR MLE-STAR-Pro-1.5","split":"all","subject_type":"system"},"harnessId":"CAIR MLE-STAR-Pro-1.5","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-82a88f0ed43374a2473f8429","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"high","value":33.33,"rawValue":"33.33 ± 0.00","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=22;column=High (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-10-10","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Famou-Agent","split":"high","subject_type":"system"},"harnessId":"Famou-Agent","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-801cc75aa5a62bbe9c0f193f","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"high","value":33.33,"rawValue":"33.33 ±  0.00","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=21;column=High (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-11-25","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"CAIR MLE-STAR-Pro-1.5","split":"high","subject_type":"system"},"harnessId":"CAIR MLE-STAR-Pro-1.5","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-6ec17d8bbbddcfd84c94f9dd","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"high","value":31.11,"rawValue":"31.11 ± 2.22","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=24;column=High (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-11-03","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"CAIR MLE-STAR-Pro-1.0","split":"high","subject_type":"system"},"harnessId":"CAIR MLE-STAR-Pro-1.0","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-1f3dcdbd514c6b963a328cc3","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"high","value":40.0,"rawValue":"40.00 ± 0.00","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=15;column=High (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-12-27","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Famou-Agent 2.0","split":"high","subject_type":"system"},"harnessId":"Famou-Agent 2.0","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-ef8626f48171acd78a5bcd43","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"lite","value":66.67,"rawValue":"66.67 ± 1.52","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=24;column=Low == Lite (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-11-03","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"CAIR MLE-STAR-Pro-1.0","split":"lite","subject_type":"system"},"harnessId":"CAIR MLE-STAR-Pro-1.0","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-d3434f176d4cad4a8aeb479b","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"lite","value":75.76,"rawValue":"75.76 ± 1.52","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=15;column=Low == Lite (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-12-27","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Famou-Agent 2.0","split":"lite","subject_type":"system"},"harnessId":"Famou-Agent 2.0","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-7d7fa4c99a8c39a3e85c2c99","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"lite","value":68.18,"rawValue":"68.18 ± 2.62","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=21;column=Low == Lite (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-11-25","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"CAIR MLE-STAR-Pro-1.5","split":"lite","subject_type":"system"},"harnessId":"CAIR MLE-STAR-Pro-1.5","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-7124d37289a43f3379edd1b0","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"lite","value":62.12,"rawValue":"62.12 ± 1.52","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=22;column=Low == Lite (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-10-10","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Famou-Agent","split":"lite","subject_type":"system"},"harnessId":"Famou-Agent","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-d7940b963c2dc7eb924cb6d5","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"medium","value":57.89,"rawValue":"57.89 ± 1.52","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=15;column=Medium (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-12-27","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Famou-Agent 2.0","split":"medium","subject_type":"system"},"harnessId":"Famou-Agent 2.0","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c1310a5429640281b0b38928","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"medium","value":25.44,"rawValue":"25.44  ±  0.88","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=24;column=Medium (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-11-03","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"CAIR MLE-STAR-Pro-1.0","split":"medium","subject_type":"system"},"harnessId":"CAIR MLE-STAR-Pro-1.0","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-96278a3cdcfdf2f7b980687e","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"medium","value":34.21,"rawValue":"34.21 ± 1.52","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=21;column=Medium (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-11-25","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"CAIR MLE-STAR-Pro-1.5","split":"medium","subject_type":"system"},"harnessId":"CAIR MLE-STAR-Pro-1.5","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-2d29eadd6b0c537bd2a118a8","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini-2.5-Pro","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"medium","value":36.84,"rawValue":"36.84 ± 1.52","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=22;column=Medium (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-10-10","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Famou-Agent","split":"medium","subject_type":"system"},"harnessId":"Famou-Agent","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":4},{"id":"pub-a644d7cb33f963068aab6ae8","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"all","value":35.11,"rawValue":"35.11 ± 0.44","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=26;column=All (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-09-26","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"R&D-Agent","split":"all","subject_type":"system"},"harnessId":"R&D-Agent","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-c7a4068bf711334639b2092c","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"high","value":22.22,"rawValue":"22.22 ± 2.22","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=26;column=High (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-09-26","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"R&D-Agent","split":"high","subject_type":"system"},"harnessId":"R&D-Agent","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-672ba632ed6e4a7c00c33717","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"lite","value":68.18,"rawValue":"68.18 ± 2.62","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=26;column=Low == Lite (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-09-26","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"R&D-Agent","split":"lite","subject_type":"system"},"harnessId":"R&D-Agent","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f5cd790fdb9657204c5652ef","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"medium","value":21.05,"rawValue":"21.05 ± 1.52","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=26;column=Medium (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-09-26","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"R&D-Agent","split":"medium","subject_type":"system"},"harnessId":"R&D-Agent","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0f3c799f6f718c0ba9064a52","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5 (low verbosity/effort)[^2]","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"all","value":39.56,"rawValue":"39.56 ± 0.44[^3]","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=23;column=All (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-10-06","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Operand ensemble","split":"all","subject_type":"system"},"harnessId":"Operand ensemble","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-dff066a65d8c6119d3911d67","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5 (low verbosity/effort)[^2]","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"high","value":20.0,"rawValue":"20.00 ± 0.00[^3]","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=23;column=High (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-10-06","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Operand ensemble","split":"high","subject_type":"system"},"harnessId":"Operand ensemble","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3b38c208f8e46aa6c75565a4","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5 (low verbosity/effort)[^2]","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"lite","value":63.64,"rawValue":"63.64 ± 0.00","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=23;column=Low == Lite (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-10-06","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Operand ensemble","split":"lite","subject_type":"system"},"harnessId":"Operand ensemble","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-68fc27bc98b161b978d4b55d","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5 (low verbosity/effort)[^2]","benchmarkId":"mle-bench","benchmarkName":"MLE-bench","benchmarkVersionHint":"mle-bench@public","metricId":"medium","value":33.33,"rawValue":"33.33 ± 0.88[^3]","unit":"percent","evidenceLevel":"A","sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourcePageUrl":"https://github.com/openai/mle-bench/","sourceApiUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=23;column=Medium (%)","retrievedAt":"2026-08-27T08:26:35Z","observedAt":"2025-10-06","payloadSha256":"586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","protocol":{"harness":"Operand ensemble","split":"medium","subject_type":"system"},"harnessId":"Operand ensemble","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-eafa7183bafcda3ee3860ed3","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"MiniMaxAI/MiniMax-M2.1","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":88.0,"rawValue":88,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=1;row=0","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/dd/c9/ddc99cec-f5d3-4cc4-9cc5-bb534377b5f6.json"},{"id":"pub-4fba15c8e5d2de0dadb9f794","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMaxAI/MiniMax-M2.5","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":80.1,"rawValue":80.1,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=41;row=40","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/c5/4c/c54c4ee8-ff99-4cda-a81f-a2e3a4347fb8.json"},{"id":"pub-015f67a892d2eed941ad6155","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"Qwen/Qwen3.5-122B-A10B","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":86.7,"rawValue":86.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=8;row=7","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/af/3c/af3c554a-1eb9-4a79-80dd-678c7aebf0aa.json"},{"id":"pub-4ce07650bd2e8098069cbc50","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"Qwen/Qwen3.5-27B","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":86.1,"rawValue":86.1,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=12;row=11","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/15/df/15df0778-3ca6-4ff9-9b34-41f74dc4ced4.json"},{"id":"pub-d206c11ca3ce520cb7eee073","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"Qwen/Qwen3.5-397B-A17B","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":87.8,"rawValue":87.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=3;row=2","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-397B-A17B"},{"id":"pub-68e5924c5958bc2ca74eedda","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"Qwen/Qwen3.6-27B","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":86.2,"rawValue":86.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=10;row=9","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu-pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-27B"},{"id":"pub-c611ab70e0cfe6e45081e58d","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"Qwen/Qwen3.6-35B-A3B","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":85.2,"rawValue":85.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=14;row=13","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu-pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-35B-A3B"},{"id":"pub-ce1409fb230d10f29b1439b7","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"XiaomiMiMo/MiMo-V2.5-Pro","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":68.5,"rawValue":68.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=51;row=50","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu-pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro"},{"id":"pub-f7415149b45ba0686e1edfe9","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"deepseek-ai/DeepSeek-R1-0528","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":85.0,"rawValue":85,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=15;row=14","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu-pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-R1-0528"},{"id":"pub-fe041462536506b0150f8bf5","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-ai/DeepSeek-V4-Flash","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":86.4,"rawValue":86.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=9;row=8","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu-pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash"},{"id":"pub-1bfa1c78530168b18a7bd253","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-ai/DeepSeek-V4-Pro","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":87.5,"rawValue":87.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=4;row=3","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu-pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro"},{"id":"pub-57f1e3696a818675154c199c","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"moonshotai/Kimi-K2.5","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":87.1,"rawValue":87.1,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=5;row=4","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.5"},{"id":"pub-82f3d09333ae16685189c70d","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"zai-org/GLM-5","benchmarkId":"mmlu-pro","benchmarkName":"MMLU-Pro","benchmarkVersionHint":"mmlu-pro@v1","metricId":"score","value":86.0,"rawValue":86,"unit":"percent","evidenceLevel":"C","sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=13;row=12","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","protocol":{"filename":".eval_results/mmlu_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/76/7d/767d7365-4f30-4ac0-9fc4-e0ade16a161f.json"},{"id":"pub-1c1779cd9192c641b96765f2","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"osworld","benchmarkName":"OSWorld 2.0","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"osworld-v2@2026.06.24","metricId":"Score","value":66.3,"rawValue":"66.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"os_world_external.csv:row=58;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Claude Opus 4.5","subject_type":"system"},"harnessId":"Claude Opus 4.5","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.anthropic.com/news/claude-opus-4-5"},{"id":"pub-b785ba471a5adfece1d1250a","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"osworld","benchmarkName":"OSWorld 2.0","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"osworld-v2@2026.06.24","metricId":"Score","value":42.9,"rawValue":"42.9","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"os_world_external.csv:row=12;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"claude-sonnet-4-5-20250929 (15 steps)","subject_type":"system"},"harnessId":"claude-sonnet-4-5-20250929 (15 steps)","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://os-world.github.io/"},{"id":"pub-b0bbb6213344156f693fbf72","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"osworld","benchmarkName":"OSWorld 2.0","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"osworld-v2@2026.06.24","metricId":"Score","value":58.1,"rawValue":"58.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"os_world_external.csv:row=6;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"claude-sonnet-4-5-20250929 (50 steps)","subject_type":"system"},"harnessId":"claude-sonnet-4-5-20250929 (50 steps)","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2,"evidenceUrl":"https://os-world.github.io/"},{"id":"pub-a77e5ebf1c817e7c20ec3137","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"osworld","benchmarkName":"OSWorld 2.0","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"osworld-v2@2026.06.24","metricId":"Score","value":62.9,"rawValue":"62.9","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"os_world_external.csv:row=5;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"claude-sonnet-4-5-20250929 (100 steps)","subject_type":"system"},"harnessId":"claude-sonnet-4-5-20250929 (100 steps)","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3,"evidenceUrl":"https://os-world.github.io/"},{"id":"pub-d14d14cff8dcec07907876bc","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"osworld","benchmarkName":"OSWorld 2.0","benchmarkVersion":"2026-02-17","benchmarkVersionHint":"osworld-v2@2026.06.24","metricId":"Score","value":72.1,"rawValue":"72.1","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"os_world_external.csv:row=3;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"claude-sonnet-4-6 (100 steps)","subject_type":"system"},"harnessId":"claude-sonnet-4-6 (100 steps)","qualityFlags":["inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1,"evidenceUrl":"https://os-world.github.io/"},{"id":"pub-6f83646fcb0f5ed105b33efa","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"osworld","benchmarkName":"OSWorld 2.0","benchmarkVersion":"2026-01-27","benchmarkVersionHint":"osworld-v2@2026.06.24","metricId":"Score","value":63.3,"rawValue":"63.3","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"os_world_external.csv:row=59;column=Score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":"Kimi-K2.5","subject_type":"system"},"harnessId":"Kimi-K2.5","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unit_inferred_from_column"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://os-world.github.io/"},{"id":"pub-736b6b0f079dbfb6bebc50be","canonicalModelId":"deepseek/r1@0528","modelName":"DeepSeek-R1-0528","modelRef":"DeepSeek-R1-0528","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-05-28","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":27.400000000000002,"rawValue":"0.274","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=78;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-0ff91d42dd0aacc4e98b3f34","canonicalModelId":"anthropic/claude-fable-5@2026-06-09","modelName":"Claude Fable 5","modelRef":"claude-fable-5_xhigh","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-06-09","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":68.30000000000001,"rawValue":"0.683","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=28;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3722ac0dcbfccae4a48645bc","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"claude-haiku-4-5-20251001_32K","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-10-15","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":5.8999999999999995,"rawValue":"0.059","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=68;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-10-15","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e28065e8e5a75e43b1b12e39","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101_32K","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":41.8,"rawValue":"0.418","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=70;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-c2b24b0ec9274d4b42ae1196","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":43.1,"rawValue":"0.431","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=51;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a057096f9c8b853ebfeb093f","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_32K","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":46.4859437751004,"rawValue":"0.464859437751004","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=50;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-74cd2107e82eb2e3ceaaef64","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":41.04104104104104,"rawValue":"0.41041041041041043","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=49;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-397fe6a492a3fcd8a453d0f6","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_xhigh","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":50.6,"rawValue":"0.506","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=40;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-383d1d884c4ca7f49c0ba970","canonicalModelId":"anthropic/claude-opus-4.8@current","modelName":"Claude Opus 4.8","modelRef":"claude-opus-4-8_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-05-28","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":39.5,"rawValue":"0.395","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=29;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-28","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-9037b7b1b67525a4190f0b19","canonicalModelId":"anthropic/claude-opus-5@2026-07-24","modelName":"Claude Opus 5","modelRef":"claude-opus-5_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-07-24","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":56.699999999999996,"rawValue":"0.567","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=13;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a0cde89a800174adabce94eb","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":13.0,"rawValue":"0.13","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=76;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-aba07499ffe4ed851b8470ac","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929_59K","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":23.599999999999998,"rawValue":"0.236","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=64;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-77e4c54b2e18893457dcf8f3","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6_32K","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-02-17","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":28.999999999999996,"rawValue":"0.29","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=47;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6cadcc6de9dc2d361f7bfb4c","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-06-30","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":22.3,"rawValue":"0.223","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=22;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-2ff8a1d9894b509dd010bf49","canonicalModelId":"anthropic/claude-sonnet-5@2026-06-30","modelName":"Claude Sonnet 5","modelRef":"claude-sonnet-5_xhigh","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-06-30","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":25.0,"rawValue":"0.25","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=21;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-30","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ed583a5959fd7e45cd2a8a6f","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-v4-flash-0731_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-07-31","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":34.66933867735471,"rawValue":"0.3466933867735471","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=11;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7c33198fdc14fb899e11f636","canonicalModelId":"deepseek/v4-pro@2026-08-13","modelName":"DeepSeek V4 Pro (0813)","modelRef":"deepseek-v4-pro-0813_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-08-13","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":55.46639919759278,"rawValue":"0.5546639919759278","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=3;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8d8814e6d20f043f8e6424e0","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-04-24","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":56.99999999999999,"rawValue":"0.57","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=25;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4256d134d49258f051dae7fc","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"fireworks/kimi-k2p5","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-01-27","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":33.900000000000006,"rawValue":"0.339","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=53;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2c6bcb4c6a6a45d4f41de93a","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-06-17","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":56.00000000000001,"rawValue":"0.56","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=77;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ec350ee762bca590808b305f","canonicalModelId":"google/gemini-3.1-pro@2026-02-19","modelName":"Gemini 3.1 Pro","modelRef":"gemini-3.1-pro-preview","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-02-19","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":77.3,"rawValue":"0.773","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=48;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-65acca8e768885b2f9617882","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":68.4,"rawValue":"0.684","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=30;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-20f139b3f51109487fde834d","canonicalModelId":"google/gemini-3.6-flash@current","modelName":"Gemini 3.6 Flash","modelRef":"gemini-3.6-flash_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-07-21","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":68.7,"rawValue":"0.687","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=12;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-21","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6ad57ffc6b0d5fcc9101af29","canonicalModelId":"google/gemini-3.7-flash@current","modelName":"Gemini 3.7 Flash","modelRef":"gemini-3.7-flash_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-08-13","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":71.2,"rawValue":"0.712","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=4;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1b314af0cdab4ca8744677e4","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"glm-4.7","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-12-22","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":31.5,"rawValue":"0.315","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=52;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-6c48e6456800b201a519629f","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-04-07","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":37.286432160804026,"rawValue":"0.37286432160804023","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=36;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-18e646005d7f34be1170d703","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-06-16","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":38.1,"rawValue":"0.381","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=23;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3cadfa09656efff3431721e9","canonicalModelId":"zai/glm-5.3@2026-08-14","modelName":"GLM-5.3","modelRef":"glm-5.3_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-08-14","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":40.2,"rawValue":"0.402","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=2;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-14","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-768372220e23479fd3c30009","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":50.6,"rawValue":"0.506","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=74;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-83b7606527fe0dc28768dfd1","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":21.0,"rawValue":"0.21","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=67;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-27b523aaa2b652ab1cb2dce7","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"gpt-5-nano-2025-08-07_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":12.2,"rawValue":"0.122","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=72;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-294f1a876fe9d4bf38c2ef8d","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":48.9,"rawValue":"0.489","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=65;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9bf2b4d75cf85cb78a54ac04","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":38.2,"rawValue":"0.382","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=59;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f356348a4f0e622514f3798c","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_low","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":34.699999999999996,"rawValue":"0.347","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=61;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-9131b6b86b7ace0b76000a63","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_medium","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":35.4,"rawValue":"0.354","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=58;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-2eb8d346a9eb2b8fc5c836a0","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_xhigh","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":38.9,"rawValue":"0.389","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=60;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-78db8af49712adde577504f0","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_xhigh","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":44.83450351053159,"rawValue":"0.4483450351053159","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=46;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-60e68f41fed5be18f7dbc649","canonicalModelId":"openai/gpt-5.4-mini@current","modelName":"GPT-5.4 mini","modelRef":"gpt-5.4-mini-2026-03-17_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":28.599999999999998,"rawValue":"0.286","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=41;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d8162f874804661ff1ba156c","canonicalModelId":"openai/gpt-5.4-nano@2026-03-17","modelName":"GPT-5.4 nano","modelRef":"gpt-5.4-nano-2026-03-17_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-03-17","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":12.0,"rawValue":"0.12","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=42;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e2eb45294f8271fdef151f5d","canonicalModelId":"openai/gpt-5.4-pro@2026-03-05","modelName":"GPT-5.4 Pro","modelRef":"gpt-5.4-pro-2026-03-05_xhigh","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":47.8,"rawValue":"0.478","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=45;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2870bdd618a8dbccbe964988","canonicalModelId":"openai/gpt-5.5@current","modelName":"GPT-5.5","modelRef":"gpt-5.5-instant","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-05-05","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":38.0,"rawValue":"0.38","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=10;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1dae7a5bd43139af9bf9fc6a","canonicalModelId":"openai/gpt-5.6-luna@2026-07-09","modelName":"GPT-5.6 Luna","modelRef":"gpt-5.6-luna_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":41.699999999999996,"rawValue":"0.417","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=18;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b659f488fbe8c09144a121be","canonicalModelId":"openai/gpt-5.6-sol@2026-07-09","modelName":"GPT-5.6 Sol","modelRef":"gpt-5.6-sol_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":71.6,"rawValue":"0.716","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=19;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-1dbed4d1386f82ed68e48818","canonicalModelId":"openai/gpt-5.6-terra@2026-07-09","modelName":"GPT-5.6 Terra","modelRef":"gpt-5.6-terra_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-07-09","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":43.1,"rawValue":"0.431","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=17;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-09","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-428afedf5a5da7fa50c1ce26","canonicalModelId":"xai/grok-4.5@2026-07-16","modelName":"Grok 4.5","modelRef":"grok-4.5_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-07-08","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":53.5,"rawValue":"0.535","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=20;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-08","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3639210223e77260d3afd8e6","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_high","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-08-12","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":53.0,"rawValue":"0.53","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=7;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-52e58a93ef9e1f4243eee16b","canonicalModelId":"xai/grok-4.6@2026-08-12","modelName":"Grok 4.6","modelRef":"grok-4.6_xhigh","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-08-12","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":54.400000000000006,"rawValue":"0.544","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=6;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a83f0007f2f0912b1d52871d","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-04-20","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":38.7,"rawValue":"0.387","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=37;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4220fcaf105855651be90c5b","canonicalModelId":"moonshot/kimi-k2.7-code@current","modelName":"Kimi K2.7 Code","modelRef":"kimi-k2.7-code","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-06-12","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":39.2,"rawValue":"0.392","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=27;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-12","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-16d06a53fb538c68bf6ef8d2","canonicalModelId":"moonshot/kimi-k3@2026-07-16","modelName":"Kimi K3","modelRef":"kimi-k3_max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-07-16","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":42.699999999999996,"rawValue":"0.427","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=14;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-07-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-ec54ec0c99c9fde5edef4769","canonicalModelId":"qwen/qwen3.5-flash@2026-02-25","modelName":"Qwen3.5 Flash","modelRef":"qwen3.5-flash","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-02-25","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":19.7592778335005,"rawValue":"0.19759277833500502","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=35;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-25","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0f9600e4462c6cbb651a8f1b","canonicalModelId":"qwen/qwen3.5-plus@2026-02-16","modelName":"Qwen3.5 Plus","modelRef":"qwen3.5-plus","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-02-16","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":26.00401606425703,"rawValue":"0.2600401606425703","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=32;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2ac59fb8697af9dbba7c791f","canonicalModelId":"qwen/qwen3.6-flash@2026-04-27","modelName":"Qwen3.6 Flash","modelRef":"qwen3.6-flash","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-04-27","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":21.163490471414242,"rawValue":"0.21163490471414242","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=34;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-711c24e08dc351d6b4fd7143","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-04-20","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":56.92771084337349,"rawValue":"0.5692771084337349","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=31;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-fe92a415d9b118e1ec1083e9","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-03-31","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":49.09819639278557,"rawValue":"0.4909819639278557","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=33;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-67fb059c049115db34050519","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":58.517034068136276,"rawValue":"0.5851703406813628","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=26;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5ba3e038716f83bce20485c4","canonicalModelId":"qwen/qwen3.8-max@2026-08-02","modelName":"Qwen3.8 Max","modelRef":"qwen3.8-max_xhigh","benchmarkId":"simpleqa","benchmarkName":"SimpleQA","benchmarkVersion":"2026-08-02","benchmarkVersionHint":"simpleqa@v1","metricId":"mean_score","value":46.292585170340686,"rawValue":"0.46292585170340683","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simpleqa_verified.csv:row=9;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-08-02","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-c23eb2217cb9f8cf1d7f71e5","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus","benchmarkId":"swebench-bash-only","metricId":"resolved","value":67.6,"rawValue":67.6,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[16]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-02","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3b2c584f5ebd8c8caeb2c55a","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku","benchmarkId":"swebench-bash-only","metricId":"resolved","value":66.6,"rawValue":66.6,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[17]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3348e47773699ff4c046b6ca","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"Claude 4.5 Opus","benchmarkId":"swebench-bash-only","metricId":"resolved","value":76.8,"rawValue":76.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[0]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-885195423c2d9360149f63a8","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"Claude 4.5 Opus","benchmarkId":"swebench-bash-only","metricId":"resolved","value":74.4,"rawValue":74.4,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[4]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-11-24","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-4c214e724420dcd6354e21e7","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet","benchmarkId":"swebench-bash-only","metricId":"resolved","value":71.4,"rawValue":71.4,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[10]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5bbc2f607972f47610fad570","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet","benchmarkId":"swebench-bash-only","metricId":"resolved","value":70.6,"rawValue":70.6,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[12]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-09-29","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-653ea4df2110cb09ad8431de","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"Claude 4.6 Opus","benchmarkId":"swebench-bash-only","metricId":"resolved","value":75.6,"rawValue":75.6,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[3]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7eca0998b99d66e40564340f","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"DeepSeek V3.2","benchmarkId":"swebench-bash-only","metricId":"resolved","value":70.0,"rawValue":70.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[13]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["not_independently_checked"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-270e39ae065a85e897f8505c","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"DeepSeek V3.2 Reasoner","benchmarkId":"swebench-bash-only","metricId":"resolved","value":60.0,"rawValue":60.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[24]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-12-01","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["curated_model_alias"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-8eff87b3391725b0b2cc81be","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"GLM 5","benchmarkId":"swebench-bash-only","metricId":"resolved","value":72.8,"rawValue":72.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[7]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["heuristic_model_mapping","not_independently_checked"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-7c2a25c5f63652a82ac28dbe","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT 5","benchmarkId":"swebench-bash-only","metricId":"resolved","value":65.0,"rawValue":65.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[20]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-07","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","heuristic_model_mapping"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-4f629a205fe1df8a0fa6d072","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT 5 mini","benchmarkId":"swebench-bash-only","metricId":"resolved","value":59.8,"rawValue":59.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[25]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-07","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","heuristic_model_mapping"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-87992e477f49653ed9a8574b","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT 5 mini","benchmarkId":"swebench-bash-only","metricId":"resolved","value":56.2,"rawValue":56.2,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[28]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","heuristic_model_mapping","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-ea871f28f6d39c9013120f83","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT 5 nano","benchmarkId":"swebench-bash-only","metricId":"resolved","value":34.8,"rawValue":34.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[38]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-07","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","heuristic_model_mapping"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-6a92868a9aa5dbf33c16d52e","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT 5.1","benchmarkId":"swebench-bash-only","metricId":"resolved","value":66.0,"rawValue":66.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[19]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-11-20","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ce878bc6e4f77aa8d6ad30d8","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"GPT 5.1 Codex","benchmarkId":"swebench-bash-only","metricId":"resolved","value":66.0,"rawValue":66.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[18]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-11-24","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-ba9fd9c49dad4196ac2b055d","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"GPT 5.2","benchmarkId":"swebench-bash-only","metricId":"resolved","value":72.8,"rawValue":72.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[8]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7f7b84e82d507a3dc6508fce","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"GPT 5.2","benchmarkId":"swebench-bash-only","metricId":"resolved","value":71.8,"rawValue":71.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[9]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-12-11","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-b191347af15a05285e9d15c6","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"GPT 5.2","benchmarkId":"swebench-bash-only","metricId":"resolved","value":69.0,"rawValue":69.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[15]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-12-11","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-1d162d32d186598ea0f0a1ee","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"GPT 5.2 Codex","benchmarkId":"swebench-bash-only","metricId":"resolved","value":72.8,"rawValue":72.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[6]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-19","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-7266a3ec84cb75685113a084","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"Gemini 2.5 Flash","benchmarkId":"swebench-bash-only","metricId":"resolved","value":28.73,"rawValue":28.73,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[39]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-07-26","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-cdba0cc817486f9dea38693d","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini 2.5 Pro","benchmarkId":"swebench-bash-only","metricId":"resolved","value":53.6,"rawValue":53.6,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[33]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-07-26","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ade8d3f221c939a0dcf0271f","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"Kimi K2.5","benchmarkId":"swebench-bash-only","metricId":"resolved","value":70.8,"rawValue":70.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[11]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bb62e052b97cb4aa93db0565","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax M2.5","benchmarkId":"swebench-bash-only","metricId":"resolved","value":75.8,"rawValue":75.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[2]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["not_independently_checked"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-6b3438cd0566b5c8bad9e2e7","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder 480B/A35B Instruct","benchmarkId":"swebench-bash-only","metricId":"resolved","value":55.4,"rawValue":55.4,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[29]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-02","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"bash-only","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["curated_model_alias"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5e4368d4334eeb28539b82c0","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku","benchmarkId":"swebench-multilingual","benchmarkName":"SWE-bench Multilingual","benchmarkVersionHint":"swebench-multilingual@official","metricId":"resolved","value":64.7,"rawValue":64.7,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[1]=Multilingual;results[10]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-13","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Multilingual","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-5510550ec846e3cf711703b9","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"Claude 4.5 Opus","benchmarkId":"swebench-multilingual","benchmarkName":"SWE-bench Multilingual","benchmarkVersionHint":"swebench-multilingual@official","metricId":"resolved","value":70.7,"rawValue":70.7,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[1]=Multilingual;results[2]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-13","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Multilingual","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3e58721fcbae8e42750fd004","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet","benchmarkId":"swebench-multilingual","benchmarkName":"SWE-bench Multilingual","benchmarkVersionHint":"swebench-multilingual@official","metricId":"resolved","value":67.0,"rawValue":67.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[1]=Multilingual;results[7]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-13","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Multilingual","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-b896f746b94c10bbe028c65a","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"Claude 4.6 Opus","benchmarkId":"swebench-multilingual","benchmarkName":"SWE-bench Multilingual","benchmarkVersionHint":"swebench-multilingual@official","metricId":"resolved","value":72.0,"rawValue":72.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[1]=Multilingual;results[1]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-13","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Multilingual","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-3d6098563c5debb3ece9c6c0","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"DeepSeek V3.2","benchmarkId":"swebench-multilingual","benchmarkName":"SWE-bench Multilingual","benchmarkVersionHint":"swebench-multilingual@official","metricId":"resolved","value":59.0,"rawValue":59.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[1]=Multilingual;results[11]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-13","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Multilingual","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3759d4f70c9feb36ba2413ea","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"GLM 5","benchmarkId":"swebench-multilingual","benchmarkName":"SWE-bench Multilingual","benchmarkVersionHint":"swebench-multilingual@official","metricId":"resolved","value":69.7,"rawValue":69.7,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[1]=Multilingual;results[3]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-13","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Multilingual","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-beec14f752e63c902e82cb94","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT 5 mini","benchmarkId":"swebench-multilingual","benchmarkName":"SWE-bench Multilingual","benchmarkVersionHint":"swebench-multilingual@official","metricId":"resolved","value":39.7,"rawValue":39.7,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[1]=Multilingual;results[12]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-13","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Multilingual","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-d343190bcd73126efae42d14","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"GPT 5.2","benchmarkId":"swebench-multilingual","benchmarkName":"SWE-bench Multilingual","benchmarkVersionHint":"swebench-multilingual@official","metricId":"resolved","value":66.7,"rawValue":66.7,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[1]=Multilingual;results[8]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-13","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Multilingual","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-43d23ba49dcce425e7d6f67a","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"GPT 5.2 Codex","benchmarkId":"swebench-multilingual","benchmarkName":"SWE-bench Multilingual","benchmarkVersionHint":"swebench-multilingual@official","metricId":"resolved","value":66.3,"rawValue":66.3,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[1]=Multilingual;results[9]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-20","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Multilingual","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a4152c652f834029fea3e122","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"Kimi K2.5","benchmarkId":"swebench-multilingual","benchmarkName":"SWE-bench Multilingual","benchmarkVersionHint":"swebench-multilingual@official","metricId":"resolved","value":67.3,"rawValue":67.3,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[1]=Multilingual;results[6]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-13","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Multilingual","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-634fbcab24bf3baced17ad91","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax M2.5","benchmarkId":"swebench-multilingual","benchmarkName":"SWE-bench Multilingual","benchmarkVersionHint":"swebench-multilingual@official","metricId":"resolved","value":68.3,"rawValue":68.3,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[1]=Multilingual;results[5]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-16","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Multilingual","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-7c3c43c7bd6f107e96855194","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"MiniMaxAI/MiniMax-M2.1","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":36.81,"rawValue":36.81,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=38;row=37","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe_bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://scale.com/leaderboard/swe_bench_pro_public"},{"id":"pub-99fa41299706c19771544a27","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMaxAI/MiniMax-M2.5","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":55.4,"rawValue":55.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=21;row=20","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe_bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M2.5"},{"id":"pub-612bc9625dcb1142254860a0","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMaxAI/MiniMax-M2.7","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":56.2,"rawValue":56.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=18;row=17","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe-bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M2.7"},{"id":"pub-983a6cab53e0bb063950db63","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMaxAI/MiniMax-M3","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":59.0,"rawValue":59,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=10;row=9","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/minimax-m3.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M3"},{"id":"pub-9a626a7da4408eb8b9c49c44","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen/Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":38.7,"rawValue":38.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=37;row=36","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe_bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://scale.com/leaderboard/swe_bench_pro_public"},{"id":"pub-523dc5f8ccf5894f61ac135b","canonicalModelId":"qwen/qwen3-coder-next@2026-02-03","modelName":"Qwen3-Coder-Next","modelRef":"Qwen/Qwen3-Coder-Next","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":44.3,"rawValue":44.3,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=34;row=33","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/.eval_results/swe_bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/papers/2603.00729"},{"id":"pub-011e3d0118c678353230c777","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"Qwen/Qwen3.6-27B","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":53.5,"rawValue":53.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=24;row=23","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe-bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-27B"},{"id":"pub-7b005586b943e6b046d94393","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"Qwen/Qwen3.6-35B-A3B","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":49.5,"rawValue":49.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=28;row=27","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe-bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-35B-A3B"},{"id":"pub-fe4a28fdd63b09fcbb3d29f9","canonicalModelId":"qwen/qwen3.8-2.4t-a95b@2026-08-12","modelName":"Qwen3.8 2.4T A95B","modelRef":"Qwen/Qwen3.8-2.4T-A95B","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":67.7,"rawValue":67.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=1;row=0","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe-bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B"},{"id":"pub-1d3de0ab7d280e34e97ee4e5","canonicalModelId":"qwen/qwen3.8-27b@2026-08-14","modelName":"Qwen3.8 27B","modelRef":"Qwen/Qwen3.8-27B","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":61.7,"rawValue":61.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=6;row=5","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/Qwen3.8-27B.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.8-27B"},{"id":"pub-3d51674bf01108a25310aa3b","canonicalModelId":"qwen/qwen3.8-flash-next@2026-08-26","modelName":"Qwen3.8 Flash-Next","modelRef":"Qwen/Qwen3.8-Flash-Next","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":62.5,"rawValue":62.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=3;row=2","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/Qwen3.8-Flash-Next.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.8-Flash-Next"},{"id":"pub-f56a19a2a86da9b25e49e0ff","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"XiaomiMiMo/MiMo-V2.5","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":56.1,"rawValue":56.1,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=19;row=18","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe-bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5"},{"id":"pub-d0cdb142b1beda4802d22526","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"XiaomiMiMo/MiMo-V2.5-Pro","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":57.2,"rawValue":57.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=15;row=14","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe-bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro"},{"id":"pub-de19cb17291d489c274b0765","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-ai/DeepSeek-V4-Pro","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":55.4,"rawValue":55.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=22;row=21","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe-bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro"},{"id":"pub-8bdbb0fe775058a4d3683096","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"moonshotai/Kimi-K2.5","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":50.7,"rawValue":50.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=26;row=25","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe_bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.5"},{"id":"pub-2508cbed93aa0ecdca7e9b6d","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"moonshotai/Kimi-K2.6","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":58.6,"rawValue":58.6,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=11;row=10","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe_bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6"},{"id":"pub-94a68adea834cbedfe447a6c","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"zai-org/GLM-5.1","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":58.4,"rawValue":58.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=13;row=12","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe_bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.1"},{"id":"pub-69fd23cf3d3dcd826e75dc54","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"zai-org/GLM-5.2","benchmarkId":"swebench-pro","benchmarkName":"SWE-bench Pro","benchmarkVersionHint":"swebench-pro@v1","metricId":"score","value":62.1,"rawValue":62.1,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","sourceLocator":"rank=5;row=4","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","protocol":{"filename":".eval_results/swe-bench_pro.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.2"},{"id":"pub-66240617d9da857d880df8d3","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"Claude 4.5 Opus","benchmarkId":"swebench-test","metricId":"resolved","value":52.62,"rawValue":52.62,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[2]=Test;results[0]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-12-19","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":false,"harness":"Sonar Foundation Agent","leaderboard_variant":"Test","scaffold":"Sonar Foundation Agent","subject_type":"system"},"harnessId":"Sonar Foundation Agent","qualityFlags":["agent_system_score","curated_model_alias","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2258bd076fb487edb11d9dfe","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":73.2,"rawValue":73.2,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[22]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-05-22","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":false,"harness":"Tools","leaderboard_variant":"Verified","scaffold":"Tools","subject_type":"system"},"harnessId":"Tools","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-15b826a950a00a98bfe58813","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"Claude 4 Opus","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":67.6,"rawValue":67.6,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[51]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-02","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-2ea9de70b743b535c0706af2","canonicalModelId":"anthropic/claude-haiku-4.5@2025-10-01","modelName":"Claude Haiku 4.5","modelRef":"Claude 4.5 Haiku","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":66.6,"rawValue":66.6,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[53]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2f02ba214edf8933a99e29ff","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"Claude 4.5 Opus","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":79.2,"rawValue":79.2,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[0]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-12-15","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":false,"harness":"live-SWE-agent","leaderboard_variant":"Verified","scaffold":"live-SWE-agent","subject_type":"system"},"harnessId":"live-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f03a7be30416643d8f320d26","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"Claude 4.5 Opus","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":79.2,"rawValue":79.2,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[1]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-12-05","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":false,"harness":"Sonar Foundation Agent","leaderboard_variant":"Verified","scaffold":"Sonar Foundation Agent","subject_type":"system"},"harnessId":"Sonar Foundation Agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-f392f94d8f5e984fedf8babb","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"Claude 4.5 Opus","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":76.8,"rawValue":76.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[6]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-ef34c0daeba0ed5a963d660a","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"Claude 4.5 Opus","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":74.4,"rawValue":74.4,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[19]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-11-24","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":4},{"id":"pub-01c961013427f3f89b07dc5d","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":74.8,"rawValue":74.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[14]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-11-03","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":false,"harness":"Sonar Foundation Agent","leaderboard_variant":"Verified","scaffold":"Sonar Foundation Agent","subject_type":"system"},"harnessId":"Sonar Foundation Agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2a1d1d637a87a9db499bec4e","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":71.4,"rawValue":71.4,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[30]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-469352f214bd034b852326ec","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"Claude 4.5 Sonnet","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":70.6,"rawValue":70.6,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[39]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-09-29","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-55098ff346ec5bbffa3bd24f","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"Claude 4.6 Opus","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":75.6,"rawValue":75.6,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[11]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-343c5dbd13c27471c0ca5def","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"DeepSeek V3.2","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":70.0,"rawValue":70.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[45]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-3203719b90069e2e81ca76d2","canonicalModelId":"deepseek/v3.2@2025-12-01","modelName":"DeepSeek V3.2","modelRef":"DeepSeek V3.2 Reasoner","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":60.0,"rawValue":60.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[77]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-12-01","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-faafdc825432fe00b9e1b615","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"GLM 5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":72.8,"rawValue":72.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[25]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a0398be393f7a1f1210de35f","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT 5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":65.0,"rawValue":65.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[61]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-07","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":4},{"id":"pub-da035b58d3d556301d936aa4","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT 5 mini","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":59.8,"rawValue":59.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[78]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-07","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-3df2cfdc1b0908e10919bbc9","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"GPT 5 mini","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":56.2,"rawValue":56.2,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[88]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","heuristic_model_mapping","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-5b4f12f60b7d7b1afb2aee2a","canonicalModelId":"openai/gpt-5-nano@2025-08-07","modelName":"GPT-5 nano","modelRef":"GPT 5 nano","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":34.8,"rawValue":34.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[146]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-07","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-a2b95f664ba0dfccf23a0381","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"GPT 5.1","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":66.0,"rawValue":66.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[56]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-11-20","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-727fc5419de71dd9170cfdbc","canonicalModelId":"openai/gpt-5.1-codex@2025-11-12","modelName":"GPT-5.1 Codex","modelRef":"GPT 5.1 Codex","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":66.0,"rawValue":66.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[55]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-11-24","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-47fb256a8486d8914e9735a2","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"GPT 5.2","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":72.8,"rawValue":72.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[26]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-bd1260ba79ff4e833f60c7c6","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"GPT 5.2","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":71.8,"rawValue":71.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[29]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-12-11","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-ab2846599cba41047052b234","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"GPT 5.2","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":69.0,"rawValue":69.0,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[48]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-12-11","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":3},{"id":"pub-fe0156c1df5b045d06f5ab75","canonicalModelId":"openai/gpt-5.2-codex@2025-12-18","modelName":"GPT-5.2 Codex","modelRef":"GPT 5.2 Codex","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":72.8,"rawValue":72.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[24]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-19","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d929351608680f75f9b2d521","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":74.4,"rawValue":74.4,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[18]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-10-15","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":false,"harness":"Prometheus-v1.2.1","leaderboard_variant":"Verified","scaffold":"Prometheus-v1.2.1","subject_type":"system"},"harnessId":"Prometheus-v1.2.1","qualityFlags":["agent_system_score","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-94f81ac2d6c5cf97bbce25bc","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":71.8,"rawValue":71.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[28]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-07","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"OpenHands","leaderboard_variant":"Verified","scaffold":"OpenHands","subject_type":"system"},"harnessId":"OpenHands","qualityFlags":["agent_system_score","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":2},{"id":"pub-80ab4d855d41f4602922a535","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"GPT-5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":71.2,"rawValue":71.2,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[31]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-09-29","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":false,"harness":"Prometheus-v1.2","leaderboard_variant":"Verified","scaffold":"Prometheus-v1.2","subject_type":"system"},"harnessId":"Prometheus-v1.2","qualityFlags":["agent_system_score","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":3},{"id":"pub-258bb95005f11c2af16076d4","canonicalModelId":"google/gemini-2.5-flash@current","modelName":"Gemini 2.5 Flash","modelRef":"Gemini 2.5 Flash","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":28.73,"rawValue":28.73,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[154]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-07-26","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-b81068f5395da5f7062fed67","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"Gemini 2.5 Pro","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":53.6,"rawValue":53.6,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[97]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-07-26","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"system"},"harnessId":"mini-SWE-agent","qualityFlags":["agent_system_score","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"system","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-bf73363fa04ba61329604ef4","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"Kimi K2.5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":70.8,"rawValue":70.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[37]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-826c28218f2ba606903c3a23","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMax M2.5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":75.8,"rawValue":75.8,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[9]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2026-02-17","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":null,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-f631fbea42a70e26423d75cb","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"MiniMaxAI/MiniMax-M2.1","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":74.0,"rawValue":74,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=28;row=27","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M2.1"},{"id":"pub-132ffede475084ac852445fb","canonicalModelId":"minimax/m2.5@2026-02","modelName":"MiniMax M2.5","modelRef":"MiniMaxAI/MiniMax-M2.5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":75.8,"rawValue":75.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=21;row=20","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://www.swebench.com/"},{"id":"pub-c81797af459ec24e92abd2ef","canonicalModelId":"minimax/m3@2026-05-31","modelName":"MiniMax M3","modelRef":"MiniMaxAI/MiniMax-M3","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":80.5,"rawValue":80.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=6;row=5","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/minimax-m3.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M3"},{"id":"pub-43feee1dab4982c4162b7cb0","canonicalModelId":"qwen/qwen3-coder-next@2026-02-03","modelName":"Qwen3-Coder-Next","modelRef":"Qwen/Qwen3-Coder-Next","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":70.6,"rawValue":70.6,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=40;row=39","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/papers/2603.00729"},{"id":"pub-fd365c4f799b5c6338fbcc64","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"Qwen/Qwen3.5-122B-A10B","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":72.0,"rawValue":72,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=35;row=34","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-122B-A10B"},{"id":"pub-9c3fcaf6076c521eeffe7c43","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"Qwen/Qwen3.5-27B","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":72.4,"rawValue":72.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=33;row=32","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-27B"},{"id":"pub-2f4352e2ac96371ae790794d","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"Qwen/Qwen3.5-397B-A17B","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":76.4,"rawValue":76.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=19;row=18","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-397B-A17B"},{"id":"pub-610bb1943691a735b07aea80","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"Qwen/Qwen3.6-27B","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":77.2,"rawValue":77.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=18;row=17","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe-bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-27B"},{"id":"pub-e4540f62b87243401968b6d0","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"Qwen/Qwen3.6-35B-A3B","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":73.4,"rawValue":73.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=31;row=30","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe-bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-35B-A3B"},{"id":"pub-b70b4a9d3726dba75675e344","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder 480B/A35B Instruct","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":55.4,"rawValue":55.4,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[90]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-02","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":true,"harness":"mini-SWE-agent","leaderboard_variant":"Verified","scaffold":"mini-SWE-agent","subject_type":"model"},"harnessId":"mini-SWE-agent","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint"],"subjectType":"system","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-38fc0d26e0c6605142dc284b","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"resolved","value":69.6,"rawValue":69.6,"unit":"percent","evidenceLevel":"A","sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourcePageUrl":"https://www.swebench.com/","sourceApiUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[46]","retrievedAt":"2026-08-27T04:57:09Z","observedAt":"2025-08-05","payloadSha256":"fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006","protocol":{"checked":false,"harness":"OpenHands","leaderboard_variant":"Verified","scaffold":"OpenHands","subject_type":"model"},"harnessId":"OpenHands","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","not_independently_checked"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-287d106fa990a79bb917df85","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"XiaomiMiMo/MiMo-V2.5-Pro","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":78.9,"rawValue":78.9,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=12;row=11","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe-bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro"},{"id":"pub-441234e156b65371e01dbeeb","canonicalModelId":"anthropic/claude-opus@4","modelName":"Claude Opus 4","modelRef":"claude-opus-4-20250514","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2025-05-22","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":70.66115702479338,"rawValue":"0.7066115702479339","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=31;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-05-22","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a0d0fe0a2ec644ab2c6f86de","canonicalModelId":"anthropic/claude-opus-4.5@2025-11-01","modelName":"Claude Opus 4.5","modelRef":"claude-opus-4-5-20251101","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2025-11-24","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":76.65289256198346,"rawValue":"0.7665289256198347","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=32;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-e3e3a2d0a7f5304cf01f07d8","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":78.71900826446281,"rawValue":"0.7871900826446281","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=17;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-2aa011c75babc1b951800ec8","canonicalModelId":"anthropic/claude-opus-4.6@2026-02-05","modelName":"Claude Opus 4.6","modelRef":"claude-opus-4-6","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":75.6198347107438,"rawValue":"0.756198347107438","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=28;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-8f826623cbc115e64d6a920c","canonicalModelId":"anthropic/claude-opus-4.7@current","modelName":"Claude Opus 4.7","modelRef":"claude-opus-4-7_max","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-04-16","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":83.47107438016529,"rawValue":"0.8347107438016529","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=11;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-83fe1205a48e3005878cf562","canonicalModelId":"anthropic/claude-sonnet-4.5@2025-09-29","modelName":"Claude Sonnet 4.5","modelRef":"claude-sonnet-4-5-20250929","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2025-09-29","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":71.28099173553719,"rawValue":"0.7128099173553719","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=33;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-09-29","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4246b88f75a7a0f774add2b3","canonicalModelId":"anthropic/claude-sonnet-4.6@current","modelName":"Claude Sonnet 4.6","modelRef":"claude-sonnet-4-6","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-02-17","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":75.20661157024794,"rawValue":"0.7520661157024794","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=15;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-aaf63773eb743c9029b1b42f","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-ai/DeepSeek-V4-Flash","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":79.0,"rawValue":79,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=10;row=9","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe-bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash"},{"id":"pub-5d4ce2443f61a1272cd4cadc","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-ai/DeepSeek-V4-Pro","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":80.6,"rawValue":80.6,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=5;row=4","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe-bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro"},{"id":"pub-555aa442e010495a4ba6a856","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-v4-pro_max","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-04-24","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":77.63975155279503,"rawValue":"0.7763975155279503","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=4;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-24","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-84464497526161af34db3f5e","canonicalModelId":"google/gemini-2.5-pro@2025-06-05","modelName":"Gemini 2.5 Pro","modelRef":"gemini-2.5-pro","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2025-06-17","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":57.55693581780539,"rawValue":"0.5755693581780539","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=22;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-06-17","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-ac21a30467f04740a69a8e89","canonicalModelId":"google/gemini-3.5-flash@current","modelName":"Gemini 3.5 Flash","modelRef":"gemini-3.5-flash_high","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":79.33884297520662,"rawValue":"0.7933884297520661","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=5;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-1eb47b658dd7a02d9ad473bd","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"glm-5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-02-11","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":72.07792207792207,"rawValue":"0.7207792207792207","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=21;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-52c08732904570300c88ca4e","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"glm-5.1","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-04-07","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":74.17,"rawValue":"0.7417","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=7;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-e1d33c111b525aad3f139e4b","canonicalModelId":"zai/glm-5.2@2026-06-16","modelName":"GLM-5.2","modelRef":"glm-5.2_max","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-06-16","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":78.7,"rawValue":"0.787","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=2;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-06-16","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-12a208fe8351327b872e98fa","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_high","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":73.55371900826447,"rawValue":"0.7355371900826446","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=30;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2},{"id":"pub-5a090d6b131c30e9e7401ba3","canonicalModelId":"openai/gpt-5@2025-08-07","modelName":"GPT-5","modelRef":"gpt-5-2025-08-07_medium","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":71.48760330578511,"rawValue":"0.7148760330578512","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=34;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","heuristic_model_mapping","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1},{"id":"pub-b0b34132a6cceb51b74781b9","canonicalModelId":"openai/gpt-5-mini@2025-08-07","modelName":"GPT-5 mini","modelRef":"gpt-5-mini-2025-08-07_medium","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2025-08-07","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":64.6694214876033,"rawValue":"0.6466942148760331","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=36;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-08-07","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-8f1e54a95f079c1eeb6540a2","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":67.97520661157024,"rawValue":"0.6797520661157025","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=18;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-0de55f21051348b6bd9046fb","canonicalModelId":"openai/gpt-5.1@2025-11-13","modelName":"GPT-5.1","modelRef":"gpt-5.1-2025-11-13_high","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2025-11-13","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":65.9090909090909,"rawValue":"0.6590909090909091","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=20;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-11-13","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":2},{"id":"pub-a03ff70242bea3f56f5666fc","canonicalModelId":"openai/gpt-5.2@2025-12-11","modelName":"GPT-5.2","modelRef":"gpt-5.2-2025-12-11_high","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2025-12-11","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":73.7603305785124,"rawValue":"0.737603305785124","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=24;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2025-12-11","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-4c47737a12434e6b16ec8231","canonicalModelId":"openai/gpt-5.3-codex@2026-02-05","modelName":"GPT-5.3 Codex","modelRef":"gpt-5.3-codex_high","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-02-05","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":74.79338842975206,"rawValue":"0.7479338842975206","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=13;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-02-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-a1d51c870efd338cbdd496c6","canonicalModelId":"openai/gpt-5.4@current","modelName":"GPT-5.4","modelRef":"gpt-5.4-2026-03-05_high","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-03-05","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":76.85950413223141,"rawValue":"0.768595041322314","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=12;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-05","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-d7e95c8d7a50fe23977621fd","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"kimi-k2.5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-01-27","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":73.7603305785124,"rawValue":"0.737603305785124","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=19;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-01-27","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-f973a5ae46b90c602cf5cee2","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"kimi-k2.6","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-04-20","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":76.65289256198346,"rawValue":"0.7665289256198347","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=9;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-4b68987a9f33d5226493afe6","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"moonshotai/Kimi-K2.5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":70.8,"rawValue":70.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=39;row=38","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.swebench.com/"},{"id":"pub-eedc52e9754acd593d041ed7","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"moonshotai/Kimi-K2.6","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":80.2,"rawValue":80.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=7;row=6","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6"},{"id":"pub-6b6615fe11d30021084af08f","canonicalModelId":"qwen/qwen3.6-max@preview","modelName":"Qwen3.6 Max Preview","modelRef":"qwen3.6-max-preview","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-04-20","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":76.65289256198346,"rawValue":"0.7665289256198347","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=6;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-04-20","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"exact_alias","selectionRank":1},{"id":"pub-8f29d12dca1f1f98d1028e96","canonicalModelId":"qwen/qwen3.6-plus@2026-03-31","modelName":"Qwen3.6 Plus","modelRef":"qwen3.6-plus","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-03-31","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":57.85123966942148,"rawValue":"0.5785123966942148","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=8;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-03-31","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-353f75c96a41e241a96c6a4b","canonicalModelId":"qwen/qwen3.7-max@2026-05-19","modelName":"Qwen3.7 Max","modelRef":"qwen3.7-max","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersion":"2026-05-19","benchmarkVersionHint":"swebench-verified@verified","metricId":"mean_score","value":77.27272727272727,"rawValue":"0.7727272727272727","unit":"percent","evidenceLevel":"A","sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","sourcePageUrl":"https://epoch.ai/benchmarks/use-this-data","sourceApiUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"swe_bench_verified.csv:row=3;column=mean_score","retrievedAt":"2026-08-27T08:25:00Z","observedAt":"2026-05-19","payloadSha256":"30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","protocol":{"harness":null,"subject_type":"model"},"qualityFlags":["curated_model_alias","fraction_scaled_to_percent","inferred_benchmark_version_hint"],"subjectType":"model","sourceSubjectType":"model","mappingStatus":"curated_alias","selectionRank":1},{"id":"pub-23bedbd09170c27a18458c05","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/GLM-4.7","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":73.8,"rawValue":73.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=30;row=29","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-4.7"},{"id":"pub-f634e67ccde86baa0c9e7e57","canonicalModelId":"zai/glm-4.7-flash@current","modelName":"GLM-4.7 Flash","modelRef":"zai-org/GLM-4.7-Flash","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":59.2,"rawValue":59.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=59;row=58","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-4.7-Flash"},{"id":"pub-f6cd0cfcd446a60dc5434d4b","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"zai-org/GLM-5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":77.8,"rawValue":77.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=15;row=14","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5/"},{"id":"pub-bebdae5a7299c00757f819be","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"zai-org/GLM-5","benchmarkId":"swebench-verified","benchmarkName":"SWE-bench Verified","benchmarkVersionHint":"swebench-verified@verified","metricId":"score","value":72.8,"rawValue":72.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","sourceLocator":"rank=32;row=31","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","protocol":{"filename":".eval_results/swe_bench_verified.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://www.swebench.com/"},{"id":"pub-09b19fb6a8ca94e8db14a39c","canonicalModelId":"minimax/m2.1@2025-12-23","modelName":"MiniMax M2.1","modelRef":"MiniMaxAI/MiniMax-M2.1","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":29.2,"rawValue":29.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=34;row=33","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal_bench.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-0e2cafc0c4578e634f6fb919","canonicalModelId":"minimax/m2.7@2026-03-18","modelName":"MiniMax M2.7","modelRef":"MiniMaxAI/MiniMax-M2.7","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":57.0,"rawValue":57,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=10;row=9","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal-bench-2.0.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M2.7"},{"id":"pub-e84f87f42832768d2767bb0d","canonicalModelId":"qwen/qwen3-coder@current","modelName":"Qwen3-Coder 480B-A35B","modelRef":"Qwen/Qwen3-Coder-480B-A35B-Instruct","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":23.9,"rawValue":23.9,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=38;row=37","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal_bench.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-2d522fec927e539e48698248","canonicalModelId":"qwen/qwen3-coder-next@2026-02-03","modelName":"Qwen3-Coder-Next","modelRef":"Qwen/Qwen3-Coder-Next","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":36.2,"rawValue":36.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=22;row=21","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal_bench_2.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3-Coder-Next"},{"id":"pub-c36c5468c4676c1dc6a30cde","canonicalModelId":"qwen/qwen3.5-122b-a10b@2026-02-13","modelName":"Qwen3.5 122B A10B","modelRef":"Qwen/Qwen3.5-122B-A10B","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":49.4,"rawValue":49.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=17;row=16","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal-bench-2_2.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-122B-A10B"},{"id":"pub-bc71e9455400af5bbcb90d78","canonicalModelId":"qwen/qwen3.5-27b@2026-02-24","modelName":"Qwen3.5 27B","modelRef":"Qwen/Qwen3.5-27B","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":41.6,"rawValue":41.6,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=20;row=19","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal-bench-2.0_4.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-27B"},{"id":"pub-994efcc0153c2221f3fcddd5","canonicalModelId":"qwen/qwen3.5-397b-a17b@2026-02-13","modelName":"Qwen3.5 397B A17B","modelRef":"Qwen/Qwen3.5-397B-A17B","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":52.5,"rawValue":52.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=13;row=12","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminalbench_2.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.5-397B-A17B"},{"id":"pub-d107962c9d82877d516d8a62","canonicalModelId":"qwen/qwen3.6-27b@2026-04-22","modelName":"Qwen3.6 27B","modelRef":"Qwen/Qwen3.6-27B","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":59.3,"rawValue":59.3,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=9;row=8","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal-bench-2.0.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-27B"},{"id":"pub-f762ecb0f96fd55823cbd164","canonicalModelId":"qwen/qwen3.6-35b-a3b@2026-04-14","modelName":"Qwen3.6 35B A3B","modelRef":"Qwen/Qwen3.6-35B-A3B","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":51.5,"rawValue":51.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=15;row=14","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal-bench-2.0.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/Qwen/Qwen3.6-35B-A3B"},{"id":"pub-787330cefc9f3c4eb899140e","canonicalModelId":"xiaomi/mimo-v2.5@2026-04-23","modelName":"MiMo-V2.5","modelRef":"XiaomiMiMo/MiMo-V2.5","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":65.8,"rawValue":65.8,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=7;row=6","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal-bench-2.0.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5"},{"id":"pub-21d6c77a10f527526f00c859","canonicalModelId":"xiaomi/mimo-v2.5-pro@2026-04-23","modelName":"MiMo-V2.5-Pro","modelRef":"XiaomiMiMo/MiMo-V2.5-Pro","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":68.4,"rawValue":68.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=3;row=2","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal-bench-2.0.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/XiaomiMiMo/MiMo-V2.5-Pro"},{"id":"pub-a8aa0f1936041fbc4046a907","canonicalModelId":"deepseek/v4-flash@2026-07-31","modelName":"DeepSeek V4 Flash","modelRef":"deepseek-ai/DeepSeek-V4-Flash","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":56.9,"rawValue":56.9,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=11;row=10","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal-bench-2.0.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash"},{"id":"pub-6b383488a5beac3b1b56ffe9","canonicalModelId":"deepseek/v4-pro@2026-04-24","modelName":"DeepSeek V4 Pro","modelRef":"deepseek-ai/DeepSeek-V4-Pro","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":67.9,"rawValue":67.9,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=4;row=3","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal-bench-2.0.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro"},{"id":"pub-5f7c0efaf2d0252ec1ac2f87","canonicalModelId":"moonshot/kimi-k2.5@2026-01-27","modelName":"Kimi K2.5","modelRef":"moonshotai/Kimi-K2.5","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":43.2,"rawValue":43.2,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=19;row=18","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal_bench.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-be2abf33b38a4c20ee1904ee","canonicalModelId":"moonshot/kimi-k2.6@2026-04-20","modelName":"Kimi K2.6","modelRef":"moonshotai/Kimi-K2.6","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":66.7,"rawValue":66.7,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=5;row=4","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal_bench_2.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6"},{"id":"pub-fcd8dc973ecd35b433ba160c","canonicalModelId":"zai/glm-4.7@2025-12-22","modelName":"GLM-4.7","modelRef":"zai-org/GLM-4.7","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":33.4,"rawValue":33.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=26;row=25","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal_bench.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["curated_model_alias","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"curated_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-bd4d5b802924c77f8e53698c","canonicalModelId":"zai/glm-5@2026-02-12","modelName":"GLM-5","modelRef":"zai-org/GLM-5","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":52.4,"rawValue":52.4,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=14;row=13","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal_bench_2.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0"},{"id":"pub-0d91ab2703545651dcdc874c","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"zai-org/GLM-5.1","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":69.0,"rawValue":69,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=2;row=1","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal_bench_2_claudecode.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":1,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.1"},{"id":"pub-2775fcf7296963af75b0c764","canonicalModelId":"zai/glm-5.1@2026-04-07","modelName":"GLM-5.1","modelRef":"zai-org/GLM-5.1","benchmarkId":"terminal-bench","benchmarkName":"Terminal-Bench 2.1","benchmarkVersionHint":"terminal-bench@2.1","metricId":"score","value":63.5,"rawValue":63.5,"unit":"percent","evidenceLevel":"C","sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourcePageUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","sourceLocator":"rank=8;row=7","retrievedAt":"2026-08-27T06:23:04Z","payloadSha256":"a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","protocol":{"filename":".eval_results/terminal_bench_2.yaml","harness":"huggingface-eval-results","verified":false},"harnessId":"huggingface-eval-results","qualityFlags":["heuristic_model_mapping","inferred_benchmark_version_hint","unverified_submission"],"subjectType":"model","mappingStatus":"heuristic_alias","selectionRank":2,"evidenceUrl":"https://huggingface.co/zai-org/GLM-5.1"}],"publicUnmappedModels":[{"modelRef":"text-davinci-001","rowCount":75,"numericRowCount":75,"benchmarkCount":13,"benchmarkIds":["epoch-adversarial_nli_external","epoch-arc_ai2_external","epoch-bool_q_external","epoch-common_sense_qa_2_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-science_qa_external","epoch-superglue_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":75},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-01-27","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/pdf/2201.11990","sourceLocator":"adversarial_nli_external.csv:row=14;column=Score","observedAt":"2022-01-27"}]},{"modelRef":"Amazon Nova Lite","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=48;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Amazon Nova Micro","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=49;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Amazon Nova Pro","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=47;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Arctic Instruct","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=36;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Claude 2.0","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=52;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Claude 2.1","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=53;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Claude 3 Haiku (20240307)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=54;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Claude 3 Opus (20240229)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=56;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Claude 3 Sonnet (20240229)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=55;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Claude 3.5 Haiku (20241022)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=57;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Claude 3.5 Sonnet (20240620)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=58;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Claude 3.5 Sonnet (20241022)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=59;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Claude Instant 1.2","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=51;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Claude v1.3","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=50;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Command","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=60;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Command Light","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=61;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Command R","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=62;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Command R Plus","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=63;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"DBRX Instruct","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=2;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"DeepSeek LLM Chat (67B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=3;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"DeepSeek v3","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=4;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Falcon (40B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=5;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Falcon (7B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=6;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"GPT-3.5 (text-davinci-002)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=79;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"GPT-3.5 (text-davinci-003)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=78;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"GPT-3.5 Turbo (0613)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=80;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"GPT-4 (0613)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=82;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"GPT-4 Turbo (1106 preview)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=81;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"GPT-4 Turbo (2024-04-09)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=83;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"GPT-4o (2024-05-13)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=84;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"GPT-4o (2024-08-06)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=85;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"GPT-4o mini (2024-07-18)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=86;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Gemini 1.0 Pro (002)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=64;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Gemini 1.5 Flash (001)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=66;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Gemini 1.5 Flash (002)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=68;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Gemini 1.5 Pro (001)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=65;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Gemini 1.5 Pro (002)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=67;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Gemini 2.0 Flash (Experimental)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=69;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Gemma (7B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=9;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Gemma 2 Instruct (27B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=7;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Gemma 2 Instruct (9B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=8;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Jamba 1.5 Large","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=43;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Jamba 1.5 Mini","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=42;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Jamba Instruct","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=41;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Jurassic-2 Grande (17B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=39;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Jurassic-2 Jumbo (178B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=40;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"LLaMA (65B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=21;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Llama 2 (13B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=10;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Llama 2 (70B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=11;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Llama 2 (7B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=12;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Llama 3 (70B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=13;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Llama 3 (8B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=14;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Llama 3.1 Instruct Turbo (405B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=15;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Llama 3.1 Instruct Turbo (70B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=16;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Llama 3.1 Instruct Turbo (8B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=17;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Llama 3.2 Vision Instruct Turbo (11B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=18;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Llama 3.2 Vision Instruct Turbo (90B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=19;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Llama 3.3 Instruct Turbo (70B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=20;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Luminous Base (13B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=44;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Luminous Extended (30B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=45;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Luminous Supreme (70B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=46;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Mistral Instruct v0.3 (7B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=22;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Mistral Large (2402)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=75;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Mistral Large 2 (2407)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=76;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Mistral Medium (2312)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=74;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Mistral NeMo (2402)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=77;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Mistral Small (2402)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=73;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Mistral v0.1 (7B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=23;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Mixtral (8x22B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=24;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Mixtral (8x7B 32K seqlen)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=25;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"OLMo (7B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=26;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"PaLM-2 (Bison)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=70;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"PaLM-2 (Unicorn)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=71;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Palmyra X V2 (33B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=88;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Palmyra X V3 (72B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=89;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Palmyra-X-004","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=90;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Phi-2","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=27;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Phi-3 (14B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=0;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Phi-3 (7B)","rowCount":73,"numericRowCount":65,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=1;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Qwen1.5 (14B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=29;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Qwen1.5 (32B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=30;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Qwen1.5 (72B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=31;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Qwen1.5 (7B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=32;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Qwen1.5 Chat (110B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=28;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Qwen2 Instruct (72B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=33;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Qwen2.5 Instruct Turbo (72B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=34;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Qwen2.5 Instruct Turbo (7B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=35;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Solar Pro","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=87;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Yi (34B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=37;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Yi (6B)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=38;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"Yi Large (Preview)","rowCount":73,"numericRowCount":72,"benchmarkCount":11,"benchmarkIds":["helm-gsm8k","helm-legalbench","helm-math","helm-mean-win-rate","helm-medqa","helm-mmlu","helm-narrativeqa","helm-naturalquestions-closed-book","helm-naturalquestions-open-book","helm-openbookqa","helm-wmt-2014"],"sourceIds":["helm-lite"],"sourceLabels":["Stanford HELM · lite release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":73},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-01-10","examples":[{"sourceId":"helm-lite","sourceLabel":"Stanford HELM · lite release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/lite/benchmark_output/releases/v1.13.0/groups/core_scenarios.json","sourceLocator":"release=v1.13.0;table=Accuracy;row=72;column=GSM8K - EM","observedAt":"2025-01-10"}]},{"modelRef":"LLaMA-7B","rowCount":56,"numericRowCount":56,"benchmarkCount":13,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-science_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":56},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.05463","sourceLocator":"arc_ai2_external.csv:row=110;column=Challenge score","observedAt":"2023-02-24"}]},{"modelRef":"falcon-7b","rowCount":54,"numericRowCount":54,"benchmarkCount":12,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":54},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-04-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.05463","sourceLocator":"arc_ai2_external.csv:row=112;column=Challenge score","observedAt":"2023-04-24"}]},{"modelRef":"Llama-2-7b","rowCount":50,"numericRowCount":50,"benchmarkCount":13,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-science_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":50},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-07-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.05463","sourceLocator":"arc_ai2_external.csv:row=109;column=Challenge score","observedAt":"2023-07-18"}]},{"modelRef":"Undisclosed","rowCount":50,"numericRowCount":50,"benchmarkCount":4,"benchmarkIds":["swebench-lite","swebench-multimodal","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":50},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-06-10","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[10]","observedAt":"2024-12-08"}]},{"modelRef":"ark-0614c","rowCount":48,"numericRowCount":48,"benchmarkCount":1,"benchmarkIds":["agents-last-exam"],"sourceIds":["agents-last-exam"],"sourceLabels":["Agents' Last Exam · ALE-V1 leaderboard"],"sourceUrls":["https://agents-last-exam.org/api/demo/leaderboard"],"mappingStatusCounts":{"unmatched":48},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","evidenceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[104];split=full/full-spectrum;harness=openclaw_cli;model=ark-0614c;variant=unknown;metric=avg_score"}]},{"modelRef":"grok-4-3","rowCount":48,"numericRowCount":48,"benchmarkCount":1,"benchmarkIds":["agents-last-exam"],"sourceIds":["agents-last-exam"],"sourceLabels":["Agents' Last Exam · ALE-V1 leaderboard"],"sourceUrls":["https://agents-last-exam.org/api/demo/leaderboard"],"mappingStatusCounts":{"unmatched":48},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","evidenceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[111];split=full/full-spectrum;harness=grok_cli;model=grok-4-3;variant=unknown;metric=avg_score"}]},{"modelRef":"Llama-2-13b","rowCount":48,"numericRowCount":48,"benchmarkCount":13,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-science_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":48},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-07-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2311.16867","sourceLocator":"arc_ai2_external.csv:row=12;column=Challenge score","observedAt":"2023-07-18"}]},{"modelRef":"mpt-7b","rowCount":41,"numericRowCount":41,"benchmarkCount":12,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":41},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-05-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.05463","sourceLocator":"arc_ai2_external.csv:row=111;column=Challenge score","observedAt":"2023-05-05"}]},{"modelRef":"LLaMA-13B","rowCount":40,"numericRowCount":40,"benchmarkCount":13,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-science_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":40},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=26;column=Challenge score","observedAt":"2023-02-24"}]},{"modelRef":"LLaMA-33B","rowCount":39,"numericRowCount":39,"benchmarkCount":12,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":39},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=27;column=Challenge score","observedAt":"2023-02-24"}]},{"modelRef":"gemini-3-pro-preview","rowCount":38,"numericRowCount":38,"benchmarkCount":31,"benchmarkIds":["epoch-ale_bench_external","epoch-algotune_external","epoch-apex_agents_external","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-balrog_external","epoch-chess_puzzles","epoch-cl_bench_external","epoch-critpt_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-frontiermath_tier_4","epoch-gdpval_external","epoch-geobench_external","epoch-gso_external","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-proofbench_external","epoch-rli_external","epoch-simplebench_external","epoch-terminalbench_external","epoch-vending_bench_2_external","epoch-vpct_external","epoch-webdev_arena_external","epoch-weirdml_external","frontiermath","gpqa-diamond","hle","simpleqa","swebench-verified"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":38},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-02-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=32;column=Performance","observedAt":"2025-11-18"}]},{"modelRef":"Amazon Nova Lite","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=29;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Amazon Nova Micro","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=30;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Amazon Nova Premier","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=27;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Amazon Nova Pro","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=28;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Claude 3.5 Haiku (20241022)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=31;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Claude 3.5 Sonnet (20241022)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=32;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Claude 3.7 Sonnet (20250219)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=33;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Claude 4 Sonnet (20250514)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=34;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Claude 4 Sonnet (20250514, extended thinking)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=35;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"DeepSeek v3","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=7;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"GLM-4.5-Air-FP8","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=8;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"GPT-4.1 (2025-04-14)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=52;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"GPT-4.1 mini (2025-04-14)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=53;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"GPT-4.1 nano (2025-04-14)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=54;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"GPT-4o (2024-11-20)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=50;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"GPT-4o mini (2024-07-18)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=51;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Gemini 1.5 Flash (002)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=41;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Gemini 1.5 Pro (002)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=40;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Gemini 2.0 Flash","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=42;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Gemini 2.0 Flash Lite (02-05 preview)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=43;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Gemini 2.5 Flash (04-17 preview)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=45;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Gemini 2.5 Flash-Lite","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=44;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Gemini 2.5 Pro (03-25 preview)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=46;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Gemini 3 Pro (Preview)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=47;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Grok 3 Beta","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=65;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Grok 3 mini Beta","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=66;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Grok 4 (0709)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=67;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"IBM Granite 3.3 8B Instruct","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=26;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"IBM Granite 4.0 Micro","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=5;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"IBM Granite 4.0 Small","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=6;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Kimi K2 Instruct","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=11;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Llama 3.1 Instruct Turbo (405B)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=12;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Llama 3.1 Instruct Turbo (70B)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=13;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Llama 3.1 Instruct Turbo (8B)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=14;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Llama 4 Maverick (17Bx128E) Instruct FP8","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=15;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Llama 4 Scout (17Bx16E) Instruct","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=16;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Marin 8B Instruct","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=0;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Mistral Instruct v0.3 (7B)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=17;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Mistral Large (2411)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=49;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Mistral Small 3.1 (2503)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=48;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Mixtral Instruct (8x22B)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=18;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Mixtral Instruct (8x7B)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=19;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"OLMo 2 13B Instruct November 2024","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=2;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"OLMo 2 32B Instruct March 2025","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=1;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"OLMo 2 7B Instruct November 2024","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=3;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"OLMoE 1B-7B Instruct January 2025","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=4;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Palmyra Fin","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=64;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Palmyra Med","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=63;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Palmyra X5","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=62;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Palmyra-X-004","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=61;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Qwen2.5 Instruct Turbo (72B)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=20;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Qwen2.5 Instruct Turbo (7B)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=21;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Qwen3 235B A22B FP8 Throughput","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=22;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Qwen3 235B A22B Instruct 2507 FP8","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=23;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Qwen3-Next 80B A3B Thinking","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=24;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"gpt-oss-120b","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=9;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"gpt-oss-20b","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=10;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"o3 (2025-04-16)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=59;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"o4-mini (2025-04-16)","rowCount":37,"numericRowCount":37,"benchmarkCount":6,"benchmarkIds":["helm-gpqa","helm-ifeval","helm-mean-score","helm-mmlu-pro","helm-omni-math","helm-wildbench"],"sourceIds":["helm-capabilities"],"sourceLabels":["Stanford HELM · capabilities release artifact"],"sourceUrls":["https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json"],"mappingStatusCounts":{"unmatched":37},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"helm-capabilities","sourceLabel":"Stanford HELM · capabilities release artifact","sourceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","evidenceUrl":"https://storage.googleapis.com/crfm-helm-public/capabilities/benchmark_output/releases/v1.15.0/groups/core_scenarios.json","sourceLocator":"release=v1.15.0;table=Accuracy;row=60;column=GPQA - COT correct","observedAt":"2025-11-24"}]},{"modelRef":"Llama-2-70b-hf","rowCount":36,"numericRowCount":36,"benchmarkCount":13,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-common_sense_qa_2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":36},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-07-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2311.16867","sourceLocator":"arc_ai2_external.csv:row=14;column=Challenge score","observedAt":"2023-07-18"}]},{"modelRef":"LLaMA-65B","rowCount":35,"numericRowCount":35,"benchmarkCount":12,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":35},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=28;column=Challenge score","observedAt":"2023-02-24"}]},{"modelRef":"gemini-3-flash-preview","rowCount":34,"numericRowCount":34,"benchmarkCount":25,"benchmarkIds":["epoch-ale_bench_external","epoch-apex_agents_external","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-balrog_external","epoch-blueprint_bench_2_external","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-frontiermath_tier_4","epoch-gdp_pdf_external","epoch-geobench_external","epoch-gso_external","epoch-otis_mock_aime_2024_2025","epoch-proofbench_external","epoch-simplebench_external","epoch-terminalbench_external","epoch-vending_bench_2_external","epoch-vpct_external","epoch-webdev_arena_external","epoch-weirdml_external","frontiermath","gpqa-diamond","simpleqa","swebench-verified"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":34},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-03-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=18;column=Performance","observedAt":"2025-12-17"}]},{"modelRef":"Multiple","rowCount":33,"numericRowCount":33,"benchmarkCount":3,"benchmarkIds":["swebench-lite","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":33},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-11-03","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[11]","observedAt":"2025-06-13"}]},{"modelRef":"falcon-40b","rowCount":31,"numericRowCount":31,"benchmarkCount":12,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":31},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-03-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2311.16867","sourceLocator":"arc_ai2_external.csv:row=16;column=Challenge score","observedAt":"2023-03-15"}]},{"modelRef":"Mistral-7B-v0.1","rowCount":30,"numericRowCount":30,"benchmarkCount":12,"benchmarkIds":["epoch-adversarial_nli_external","epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":30},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-27","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"adversarial_nli_external.csv:row=9;column=Score","observedAt":"2023-09-27"}]},{"modelRef":"GPT-4o","rowCount":30,"numericRowCount":30,"benchmarkCount":4,"benchmarkIds":["swebench-lite","swebench-multimodal","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":30},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-05-31","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[33]","observedAt":"2024-06-27"}]},{"modelRef":"Claude 3.5 Sonnet","rowCount":29,"numericRowCount":29,"benchmarkCount":4,"benchmarkIds":["swebench-lite","swebench-multimodal","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":29},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-09-11","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[14]","observedAt":"2024-11-22"}]},{"modelRef":"PaLM 540B","rowCount":28,"numericRowCount":28,"benchmarkCount":10,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":28},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-04-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2311.16867","sourceLocator":"arc_ai2_external.csv:row=10;column=Challenge score","observedAt":"2022-04-04"}]},{"modelRef":"Llama-2-34b","rowCount":26,"numericRowCount":26,"benchmarkCount":10,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":26},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-07-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2311.16867","sourceLocator":"arc_ai2_external.csv:row=13;column=Challenge score","observedAt":"2023-07-18"}]},{"modelRef":"grok-4-0709","rowCount":25,"numericRowCount":25,"benchmarkCount":20,"benchmarkIds":["aider-polyglot","epoch-apex_agents_external","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-balrog_external","epoch-chess_puzzles","epoch-deepresearchbench_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-frontiermath_tier_4","epoch-geobench_external","epoch-lech_mazur_writing_external","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-terminalbench_external","epoch-weirdml_external","frontiermath","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":25},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=14;column=Percent correct","observedAt":"2025-07-09"}]},{"modelRef":"composer-2-5","rowCount":24,"numericRowCount":24,"benchmarkCount":1,"benchmarkIds":["agents-last-exam"],"sourceIds":["agents-last-exam"],"sourceLabels":["Agents' Last Exam · ALE-V1 leaderboard"],"sourceUrls":["https://agents-last-exam.org/api/demo/leaderboard"],"mappingStatusCounts":{"unmatched":24},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","evidenceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[155];split=full/last-exam;harness=cursor_cli;model=composer-2-5;variant=unknown;metric=avg_score"}]},{"modelRef":"grok-3","rowCount":24,"numericRowCount":24,"benchmarkCount":1,"benchmarkIds":["agents-last-exam"],"sourceIds":["agents-last-exam"],"sourceLabels":["Agents' Last Exam · ALE-V1 leaderboard"],"sourceUrls":["https://agents-last-exam.org/api/demo/leaderboard"],"mappingStatusCounts":{"unmatched":24},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"agents-last-exam","sourceLabel":"Agents' Last Exam · ALE-V1 leaderboard","sourceUrl":"https://agents-last-exam.org/api/demo/leaderboard","evidenceUrl":"https://agents-last-exam.org/api/demo/leaderboard","sourceLocator":"rows[114];split=full/full-spectrum;harness=grok_cli;model=grok-3;variant=unknown;metric=avg_score"}]},{"modelRef":"Gopher (280B)","rowCount":24,"numericRowCount":24,"benchmarkCount":7,"benchmarkIds":["epoch-bool_q_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":24},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2021-12-08","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2302.13971","sourceLocator":"bool_q_external.csv:row=148;column=Score","observedAt":"2021-12-08"}]},{"modelRef":"Megatron-Turing NLG 530B","rowCount":24,"numericRowCount":24,"benchmarkCount":6,"benchmarkIds":["epoch-adversarial_nli_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-lambada_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":24},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-01-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/pdf/2201.11990","sourceLocator":"adversarial_nli_external.csv:row=15;column=Score","observedAt":"2022-01-28"}]},{"modelRef":"gemma-7b","rowCount":24,"numericRowCount":24,"benchmarkCount":12,"benchmarkIds":["epoch-adversarial_nli_external","epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":24},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-21","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"adversarial_nli_external.csv:row=10;column=Score","observedAt":"2024-02-21"}]},{"modelRef":"grok-4.3","rowCount":23,"numericRowCount":23,"benchmarkCount":23,"benchmarkIds":["livebench-amps-hard","livebench-code-completion","livebench-code-generation","livebench-connections","livebench-consecutive-events","livebench-integrals-with-game","livebench-javascript","livebench-logic-with-navigation","livebench-math-comp","livebench-olympiad","livebench-paraphrase","livebench-plot-unscrambling","livebench-python","livebench-simplify","livebench-spatial","livebench-story-generation","livebench-summarize","livebench-tablejoin","livebench-tablereformat","livebench-theory-of-mind","livebench-typescript","livebench-typos","livebench-zebra-puzzle"],"sourceIds":["livebench-official"],"sourceLabels":["LiveBench · official dated release table"],"sourceUrls":["https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv"],"mappingStatusCounts":{"unmatched":23},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026_06_25","examples":[{"sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","evidenceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourceLocator":"path=public/table_2026_06_25.csv;row=16;column=AMPS_Hard","observedAt":"2026_06_25"}]},{"modelRef":"grok-build-0.1","rowCount":23,"numericRowCount":23,"benchmarkCount":23,"benchmarkIds":["livebench-amps-hard","livebench-code-completion","livebench-code-generation","livebench-connections","livebench-consecutive-events","livebench-integrals-with-game","livebench-javascript","livebench-logic-with-navigation","livebench-math-comp","livebench-olympiad","livebench-paraphrase","livebench-plot-unscrambling","livebench-python","livebench-simplify","livebench-spatial","livebench-story-generation","livebench-summarize","livebench-tablejoin","livebench-tablereformat","livebench-theory-of-mind","livebench-typescript","livebench-typos","livebench-zebra-puzzle"],"sourceIds":["livebench-official"],"sourceLabels":["LiveBench · official dated release table"],"sourceUrls":["https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv"],"mappingStatusCounts":{"unmatched":23},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026_06_25","examples":[{"sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","evidenceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourceLocator":"path=public/table_2026_06_25.csv;row=18;column=AMPS_Hard","observedAt":"2026_06_25"}]},{"modelRef":"inkling-xhigh","rowCount":23,"numericRowCount":23,"benchmarkCount":23,"benchmarkIds":["livebench-amps-hard","livebench-code-completion","livebench-code-generation","livebench-connections","livebench-consecutive-events","livebench-integrals-with-game","livebench-javascript","livebench-logic-with-navigation","livebench-math-comp","livebench-olympiad","livebench-paraphrase","livebench-plot-unscrambling","livebench-python","livebench-simplify","livebench-spatial","livebench-story-generation","livebench-summarize","livebench-tablejoin","livebench-tablereformat","livebench-theory-of-mind","livebench-typescript","livebench-typos","livebench-zebra-puzzle"],"sourceIds":["livebench-official"],"sourceLabels":["LiveBench · official dated release table"],"sourceUrls":["https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv"],"mappingStatusCounts":{"unmatched":23},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026_06_25","examples":[{"sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","evidenceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourceLocator":"path=public/table_2026_06_25.csv;row=31;column=AMPS_Hard","observedAt":"2026_06_25"}]},{"modelRef":"muse-spark-1.1-xhigh","rowCount":23,"numericRowCount":23,"benchmarkCount":23,"benchmarkIds":["livebench-amps-hard","livebench-code-completion","livebench-code-generation","livebench-connections","livebench-consecutive-events","livebench-integrals-with-game","livebench-javascript","livebench-logic-with-navigation","livebench-math-comp","livebench-olympiad","livebench-paraphrase","livebench-plot-unscrambling","livebench-python","livebench-simplify","livebench-spatial","livebench-story-generation","livebench-summarize","livebench-tablejoin","livebench-tablereformat","livebench-theory-of-mind","livebench-typescript","livebench-typos","livebench-zebra-puzzle"],"sourceIds":["livebench-official"],"sourceLabels":["LiveBench · official dated release table"],"sourceUrls":["https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv"],"mappingStatusCounts":{"unmatched":23},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026_06_25","examples":[{"sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","evidenceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourceLocator":"path=public/table_2026_06_25.csv;row=29;column=AMPS_Hard","observedAt":"2026_06_25"}]},{"modelRef":"muse-spark-1.2-xhigh","rowCount":23,"numericRowCount":23,"benchmarkCount":23,"benchmarkIds":["livebench-amps-hard","livebench-code-completion","livebench-code-generation","livebench-connections","livebench-consecutive-events","livebench-integrals-with-game","livebench-javascript","livebench-logic-with-navigation","livebench-math-comp","livebench-olympiad","livebench-paraphrase","livebench-plot-unscrambling","livebench-python","livebench-simplify","livebench-spatial","livebench-story-generation","livebench-summarize","livebench-tablejoin","livebench-tablereformat","livebench-theory-of-mind","livebench-typescript","livebench-typos","livebench-zebra-puzzle"],"sourceIds":["livebench-official"],"sourceLabels":["LiveBench · official dated release table"],"sourceUrls":["https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv"],"mappingStatusCounts":{"unmatched":23},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026_06_25","examples":[{"sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","evidenceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourceLocator":"path=public/table_2026_06_25.csv;row=38;column=AMPS_Hard","observedAt":"2026_06_25"}]},{"modelRef":"ox-alpha-max","rowCount":23,"numericRowCount":23,"benchmarkCount":23,"benchmarkIds":["livebench-amps-hard","livebench-code-completion","livebench-code-generation","livebench-connections","livebench-consecutive-events","livebench-integrals-with-game","livebench-javascript","livebench-logic-with-navigation","livebench-math-comp","livebench-olympiad","livebench-paraphrase","livebench-plot-unscrambling","livebench-python","livebench-simplify","livebench-spatial","livebench-story-generation","livebench-summarize","livebench-tablejoin","livebench-tablereformat","livebench-theory-of-mind","livebench-typescript","livebench-typos","livebench-zebra-puzzle"],"sourceIds":["livebench-official"],"sourceLabels":["LiveBench · official dated release table"],"sourceUrls":["https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv"],"mappingStatusCounts":{"unmatched":23},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026_06_25","examples":[{"sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","evidenceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourceLocator":"path=public/table_2026_06_25.csv;row=44;column=AMPS_Hard","observedAt":"2026_06_25"}]},{"modelRef":"smaug-agentic","rowCount":23,"numericRowCount":23,"benchmarkCount":23,"benchmarkIds":["livebench-amps-hard","livebench-code-completion","livebench-code-generation","livebench-connections","livebench-consecutive-events","livebench-integrals-with-game","livebench-javascript","livebench-logic-with-navigation","livebench-math-comp","livebench-olympiad","livebench-paraphrase","livebench-plot-unscrambling","livebench-python","livebench-simplify","livebench-spatial","livebench-story-generation","livebench-summarize","livebench-tablejoin","livebench-tablereformat","livebench-theory-of-mind","livebench-typescript","livebench-typos","livebench-zebra-puzzle"],"sourceIds":["livebench-official"],"sourceLabels":["LiveBench · official dated release table"],"sourceUrls":["https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv"],"mappingStatusCounts":{"unmatched":23},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026_06_25","examples":[{"sourceId":"livebench-official","sourceLabel":"LiveBench · official dated release table","sourceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","evidenceUrl":"https://raw.githubusercontent.com/LiveBench/new-livebench/main/public/table_2026_06_25.csv","sourceLocator":"path=public/table_2026_06_25.csv;row=39;column=AMPS_Hard","observedAt":"2026_06_25"}]},{"modelRef":"claude-3-7-sonnet-20250219","rowCount":23,"numericRowCount":23,"benchmarkCount":19,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-gso_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-the_agent_company_external","epoch-vpct_external","frontiermath","gpqa-diamond","livebench","osworld","swebench-verified"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":23},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=57;column=Percent correct","observedAt":"2025-02-24"}]},{"modelRef":"gpt-4o-2024-11-20","rowCount":23,"numericRowCount":23,"benchmarkCount":23,"benchmarkIds":["aider-polyglot","epoch-apex_agents_external","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-critpt_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-gdpval_external","epoch-geobench_external","epoch-gso_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-the_agent_company_external","epoch-vpct_external","epoch-weirdml_external","frontiermath","gpqa-diamond","hle","livebench","swebench-verified"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":23},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-30","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=41;column=Percent correct","observedAt":"2024-12-30"}]},{"modelRef":"claude-3-5-sonnet-20241022","rowCount":21,"numericRowCount":21,"benchmarkCount":21,"benchmarkIds":["aider-polyglot","epoch-balrog_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-frontiermath_tier_4","epoch-geobench_external","epoch-gso_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-the_agent_company_external","epoch-vpct_external","epoch-weirdml_external","frontiermath","gpqa-diamond","hle","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":21},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=42;column=Percent correct","observedAt":"2025-01-17"}]},{"modelRef":"o3-2025-04-16_high","rowCount":21,"numericRowCount":21,"benchmarkCount":21,"benchmarkIds":["aider-polyglot","epoch-ale_bench_external","epoch-apex_agents_external","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-cl_bench_external","epoch-critpt_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-geobench_external","epoch-gso_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-weirdml_external","frontiermath","gpqa-diamond","hle","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":21},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=38;column=Percent correct","observedAt":"2025-06-25"}]},{"modelRef":"o4-mini-2025-04-16_high","rowCount":21,"numericRowCount":21,"benchmarkCount":21,"benchmarkIds":["aider-polyglot","epoch-ale_bench_external","epoch-algotune_external","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-critpt_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-gdpval_external","epoch-geobench_external","epoch-gso_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-weirdml_external","frontiermath","gpqa-diamond","hle","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":21},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=37;column=Percent correct","observedAt":"2025-04-16"}]},{"modelRef":"gemini-3-flash","rowCount":20,"numericRowCount":20,"benchmarkCount":4,"benchmarkIds":["arena-document","arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":20},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-25","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=33","observedAt":"2026-07-30"}]},{"modelRef":"gemini-3-pro","rowCount":20,"numericRowCount":20,"benchmarkCount":4,"benchmarkIds":["arena-document","arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":20},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-25","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=26","observedAt":"2026-07-30"}]},{"modelRef":"DeepSeek-V3","rowCount":20,"numericRowCount":20,"benchmarkCount":18,"benchmarkIds":["aider-polyglot","epoch-arc_ai2_external","epoch-bbh_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-hella_swag_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","epoch-simplebench_external","epoch-trivia_qa_external","epoch-wino_grande_external","frontiermath","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":20},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-26","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=47;column=Percent correct","observedAt":"2024-12-26"}]},{"modelRef":"o3-2025-04-16_medium","rowCount":20,"numericRowCount":20,"benchmarkCount":18,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-deepresearchbench_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-gdpval_external","epoch-geobench_external","epoch-lech_mazur_writing_external","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-vpct_external","frontiermath","gpqa-diamond","hle","osworld","swebench-verified"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":20},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=10;column=Percent correct","observedAt":"2025-04-16"}]},{"modelRef":"gpt-4o-mini-2024-07-18","rowCount":19,"numericRowCount":19,"benchmarkCount":18,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-balrog_external","epoch-bool_q_external","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-gsm8k_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-piqa_external","epoch-simplebench_external","epoch-vpct_external","epoch-weirdml_external","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":19},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-21","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=53;column=Percent correct","observedAt":"2024-12-21"}]},{"modelRef":"gemini-3-flash (thinking-minimal)","rowCount":18,"numericRowCount":18,"benchmarkCount":3,"benchmarkIds":["arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":18},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-25","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=457","observedAt":"2026-08-21"}]},{"modelRef":"gemini-3.1-flash-lite-preview","rowCount":18,"numericRowCount":18,"benchmarkCount":3,"benchmarkIds":["arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":18},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-25","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=115","observedAt":"2026-08-21"}]},{"modelRef":"gemma-4-31b","rowCount":18,"numericRowCount":18,"benchmarkCount":4,"benchmarkIds":["arena-document","arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":18},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=28","observedAt":"2026-07-30"}]},{"modelRef":"grok-4.3","rowCount":18,"numericRowCount":18,"benchmarkCount":4,"benchmarkIds":["arena-search","arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":18},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-25","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=22","observedAt":"2026-08-24"}]},{"modelRef":"claude-sonnet-4-20250514","rowCount":18,"numericRowCount":18,"benchmarkCount":16,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-geobench_external","epoch-gso_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-the_agent_company_external","epoch-vpct_external","epoch-weirdml_external","frontiermath","gpqa-diamond","osworld"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":18},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=21;column=Percent correct","observedAt":"2025-05-24"}]},{"modelRef":"gpt-4.1-2025-04-14","rowCount":18,"numericRowCount":18,"benchmarkCount":18,"benchmarkIds":["aider-polyglot","epoch-ale_bench_external","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-frontiermath_tier_4","epoch-geobench_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-weirdml_external","frontiermath","gpqa-diamond","hle","swebench-verified"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":18},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=32;column=Percent correct","observedAt":"2025-04-14"}]},{"modelRef":"gemma-4-26b-a4b","rowCount":17,"numericRowCount":17,"benchmarkCount":3,"benchmarkIds":["arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":17},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=458","observedAt":"2026-08-21"}]},{"modelRef":"grok-4.20-beta-0309-reasoning","rowCount":17,"numericRowCount":17,"benchmarkCount":4,"benchmarkIds":["arena-document","arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":17},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=32","observedAt":"2026-07-30"}]},{"modelRef":"muse-spark-1.1","rowCount":17,"numericRowCount":17,"benchmarkCount":4,"benchmarkIds":["arena-document","arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":17},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-25","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=12","observedAt":"2026-07-30"}]},{"modelRef":"Baichuan-2-13B-Base","rowCount":17,"numericRowCount":17,"benchmarkCount":9,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":17},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=24;column=Challenge score","observedAt":"2023-09-06"}]},{"modelRef":"DeepSeek-R1","rowCount":17,"numericRowCount":17,"benchmarkCount":17,"benchmarkIds":["aider-polyglot","epoch-algotune_external","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-balrog_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","epoch-simplebench_external","epoch-weirdml_external","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":17},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=46;column=Percent correct","observedAt":"2025-01-20"}]},{"modelRef":"claude-opus-4-1-20250805","rowCount":17,"numericRowCount":17,"benchmarkCount":14,"benchmarkIds":["epoch-chess_puzzles","epoch-deepresearchbench_external","epoch-epoch_capabilities_index","epoch-gdpval_external","epoch-lech_mazur_writing_external","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-terminalbench_external","epoch-vpct_external","epoch-webdev_arena_external","frontiermath","gpqa-diamond","swebench-verified"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":17},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-01-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=89;column=mean_score","observedAt":"2025-08-05"}]},{"modelRef":"mpt-30b","rowCount":17,"numericRowCount":17,"benchmarkCount":10,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":17},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-06-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2307.09288","sourceLocator":"arc_ai2_external.csv:row=51;column=Challenge score","observedAt":"2023-06-22"}]},{"modelRef":"o3-mini-2025-01-31_high","rowCount":17,"numericRowCount":17,"benchmarkCount":17,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-gso_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","epoch-simplebench_external","epoch-weirdml_external","frontiermath","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":17},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-31","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=69;column=Percent correct","observedAt":"2025-01-31"}]},{"modelRef":"grok-4-1-fast-reasoning","rowCount":16,"numericRowCount":16,"benchmarkCount":3,"benchmarkIds":["arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":16},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=126","observedAt":"2026-08-21"}]},{"modelRef":"Qwen-14B","rowCount":16,"numericRowCount":16,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-lambada_external","epoch-mmlu_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":16},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=35;column=Challenge score","observedAt":"2023-09-28"}]},{"modelRef":"Gemini-3-Pro-Preview","rowCount":16,"numericRowCount":16,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":16},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2026-02-23","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=10;column=All (%)","observedAt":"2026-02-23"}]},{"modelRef":"Inkling Small","rowCount":15,"numericRowCount":15,"benchmarkCount":4,"benchmarkIds":["arena-agent","arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":15},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-25","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=41","observedAt":"2026-08-24"}]},{"modelRef":"mistral-large-3","rowCount":15,"numericRowCount":15,"benchmarkCount":3,"benchmarkIds":["arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":15},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-25","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=514","observedAt":"2026-08-21"}]},{"modelRef":"mistral-medium-3.5","rowCount":15,"numericRowCount":15,"benchmarkCount":3,"benchmarkIds":["arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":15},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=513","observedAt":"2026-08-21"}]},{"modelRef":"claude-3-5-haiku-20241022","rowCount":15,"numericRowCount":15,"benchmarkCount":15,"benchmarkIds":["aider-polyglot","epoch-balrog_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","epoch-weirdml_external","frontiermath","gpqa-diamond","livebench","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":15},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-21","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=44;column=Percent correct","observedAt":"2024-12-21"}]},{"modelRef":"gpt-4.5-preview-2025-02-27","rowCount":15,"numericRowCount":15,"benchmarkCount":15,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-vpct_external","epoch-weirdml_external","gpqa-diamond","hle","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":15},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-27","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=55;column=Percent correct","observedAt":"2025-02-27"}]},{"modelRef":"chatgpt-4o-latest-20250326","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=502","observedAt":"2026-08-21"}]},{"modelRef":"gemini-2.5-flash-lite-preview-06-17-thinking","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=165","observedAt":"2026-08-21"}]},{"modelRef":"gemma-3-27b-it","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=178","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4.1-2025-04-14","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=151","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4.1-mini-2025-04-14","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=192","observedAt":"2026-08-21"}]},{"modelRef":"gpt-5-chat","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=134","observedAt":"2026-08-21"}]},{"modelRef":"grok-4-0709","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=123","observedAt":"2026-08-21"}]},{"modelRef":"mistral-medium-2505","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=164","observedAt":"2026-08-21"}]},{"modelRef":"mistral-medium-2508","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=515","observedAt":"2026-08-21"}]},{"modelRef":"mistral-small-2506","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=195","observedAt":"2026-08-21"}]},{"modelRef":"mistral-small-3.1-24b-instruct-2503","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=252","observedAt":"2026-08-21"}]},{"modelRef":"o3-2025-04-16","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=124","observedAt":"2026-08-21"}]},{"modelRef":"o4-mini-2025-04-16","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=185","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-vl-235b-a22b-instruct","rowCount":14,"numericRowCount":14,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=500","observedAt":"2026-08-21"}]},{"modelRef":"Mixtral-8x7B-v0.1","rowCount":14,"numericRowCount":14,"benchmarkCount":10,"benchmarkIds":["epoch-adversarial_nli_external","epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-12-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"adversarial_nli_external.csv:row=12;column=Score","observedAt":"2023-12-11"}]},{"modelRef":"Phi-3-medium-128k-instruct","rowCount":14,"numericRowCount":14,"benchmarkCount":11,"benchmarkIds":["epoch-adversarial_nli_external","epoch-arc_ai2_external","epoch-bbh_external","epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-math_level_5","epoch-mmlu_external","epoch-open_book_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"adversarial_nli_external.csv:row=7;column=Score","observedAt":"2024-04-23"}]},{"modelRef":"o4-mini-2025-04-16_medium","rowCount":14,"numericRowCount":14,"benchmarkCount":14,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-geobench_external","epoch-lech_mazur_writing_external","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-vpct_external","frontiermath","gpqa-diamond","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":14},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=154;column=Score","observedAt":"2025-04-16"}]},{"modelRef":"gemini-2.0-flash-001","rowCount":13,"numericRowCount":13,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":13},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=182","observedAt":"2026-08-21"}]},{"modelRef":"grok-4.20-multi-agent-beta-0309","rowCount":13,"numericRowCount":13,"benchmarkCount":3,"benchmarkIds":["arena-search","arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":13},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=12","observedAt":"2026-08-24"}]},{"modelRef":"Llama-3.3-70B-Instruct","rowCount":13,"numericRowCount":13,"benchmarkCount":12,"benchmarkIds":["epoch-balrog_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","epoch-simplebench_external","epoch-weirdml_external","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":13},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=22;column=Average progress","observedAt":"2024-12-06"}]},{"modelRef":"Llama-4-Maverick-17B-128E-Instruct","rowCount":13,"numericRowCount":13,"benchmarkCount":13,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-critpt_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-lech_mazur_writing_external","epoch-scicode_external","epoch-simplebench_external","epoch-spatialviz_bench_external","epoch-weirdml_external","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":13},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=26;column=Percent correct","observedAt":"2025-04-06"}]},{"modelRef":"Meta-Llama-3-8B-Instruct","rowCount":13,"numericRowCount":13,"benchmarkCount":11,"benchmarkIds":["epoch-adversarial_nli_external","epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-mmlu_external","epoch-open_book_qa_external","epoch-otis_mock_aime_2024_2025","epoch-trivia_qa_external","epoch-wino_grande_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":13},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"adversarial_nli_external.csv:row=11;column=Score","observedAt":"2024-04-18"}]},{"modelRef":"Phi-3-mini-4k-instruct","rowCount":13,"numericRowCount":13,"benchmarkCount":10,"benchmarkIds":["epoch-adversarial_nli_external","epoch-arc_ai2_external","epoch-bbh_external","epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":13},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"adversarial_nli_external.csv:row=5;column=Score","observedAt":"2024-04-23"}]},{"modelRef":"claude-3-opus-20240229","rowCount":13,"numericRowCount":13,"benchmarkCount":13,"benchmarkIds":["epoch-chess_puzzles","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-weirdml_external","epoch-wino_grande_external","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":13},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=94;column=mean_score","observedAt":"2024-02-29"}]},{"modelRef":"gemini-2.5-pro-preview-06-05","rowCount":13,"numericRowCount":13,"benchmarkCount":13,"benchmarkIds":["aider-polyglot","epoch-deepresearchbench_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-gso_external","epoch-metr_time_horizons_external","epoch-rli_external","epoch-simplebench_external","epoch-vpct_external","frontiermath","gpqa-diamond","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":13},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=17;column=Percent correct","observedAt":"2025-06-06"}]},{"modelRef":"grok-4-20","rowCount":13,"numericRowCount":13,"benchmarkCount":12,"benchmarkIds":["epoch-ale_bench_external","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-blueprint_bench_2_external","epoch-cl_bench_external","epoch-cl_bench_life_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-terminalbench_external","epoch-vending_bench_2_external","epoch-webdev_arena_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":13},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=35;column=Performance","observedAt":"2026-02-17"}]},{"modelRef":"Claude 4 Sonnet","rowCount":13,"numericRowCount":13,"benchmarkCount":4,"benchmarkIds":["swebench-bash-only","swebench-lite","swebench-multimodal","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":13},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-09-06","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[21]","observedAt":"2025-07-26"}]},{"modelRef":"gemini-2.5-flash-lite-preview-09-2025-no-thinking","rowCount":12,"numericRowCount":12,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=152","observedAt":"2026-08-21"}]},{"modelRef":"gemini-2.5-flash-preview-09-2025","rowCount":12,"numericRowCount":12,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=128","observedAt":"2026-08-21"}]},{"modelRef":"gpt-5.2-chat-latest-20260210","rowCount":12,"numericRowCount":12,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=490","observedAt":"2026-08-21"}]},{"modelRef":"llama-4-maverick-17b-128e-instruct","rowCount":12,"numericRowCount":12,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=238","observedAt":"2026-08-21"}]},{"modelRef":"muse-spark","rowCount":12,"numericRowCount":12,"benchmarkCount":3,"benchmarkIds":["arena-document","arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=document;split=latest;row_idx=23","observedAt":"2026-07-30"}]},{"modelRef":"Baichuan-2-7B-Base","rowCount":12,"numericRowCount":12,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=23;column=Challenge score","observedAt":"2023-09-20"}]},{"modelRef":"Chinchilla (70B)","rowCount":12,"numericRowCount":12,"benchmarkCount":6,"benchmarkIds":["epoch-bool_q_external","epoch-lambada_external","epoch-mmlu_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-03-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2302.13971","sourceLocator":"bool_q_external.csv:row=149;column=Score","observedAt":"2022-03-29"}]},{"modelRef":"DeepSeek-V3-0324","rowCount":12,"numericRowCount":12,"benchmarkCount":12,"benchmarkIds":["aider-polyglot","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","epoch-simplebench_external","epoch-weirdml_external","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=34;column=Percent correct","observedAt":"2025-03-24"}]},{"modelRef":"GLaM (MoE)","rowCount":12,"numericRowCount":12,"benchmarkCount":4,"benchmarkIds":["epoch-hella_swag_external","epoch-open_book_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2021-12-13","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2112.06905","sourceLocator":"hella_swag_external.csv:row=43;column=Overall accuracy","observedAt":"2021-12-13"}]},{"modelRef":"Llama-3.1-8B-Instruct","rowCount":12,"numericRowCount":12,"benchmarkCount":12,"benchmarkIds":["epoch-balrog_external","epoch-bool_q_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-piqa_external","epoch-scicode_external","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-07-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=30;column=Average progress","observedAt":"2024-07-23"}]},{"modelRef":"chatglm2-6b","rowCount":12,"numericRowCount":12,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-06-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=20;column=Challenge score","observedAt":"2023-06-24"}]},{"modelRef":"claude-3-5-sonnet-20240620","rowCount":12,"numericRowCount":12,"benchmarkCount":12,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-frontiermath_tier_4","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-vpct_external","epoch-weirdml_external","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-06-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=237;column=ECI Score","observedAt":"2024-06-20"}]},{"modelRef":"claude-opus-4-1-20250805_unknown","rowCount":12,"numericRowCount":12,"benchmarkCount":8,"benchmarkIds":["epoch-algotune_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-simplebench_external","epoch-terminalbench_external","epoch-webdev_arena_external","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=17;column=Score","observedAt":"2025-08-05"}]},{"modelRef":"falcon-180B","rowCount":12,"numericRowCount":12,"benchmarkCount":10,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2311.16867","sourceLocator":"arc_ai2_external.csv:row=17;column=Challenge score","observedAt":"2023-09-06"}]},{"modelRef":"gpt-4-0613","rowCount":12,"numericRowCount":12,"benchmarkCount":12,"benchmarkIds":["epoch-bbh_external","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-gsm8k_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-trivia_qa_external","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-06-13","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.10305","sourceLocator":"bbh_external.csv:row=23;column=Average","observedAt":"2023-06-13"}]},{"modelRef":"gpt-4.1-mini-2025-04-14","rowCount":12,"numericRowCount":12,"benchmarkCount":12,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","epoch-weirdml_external","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=27;column=Percent correct","observedAt":"2025-04-14"}]},{"modelRef":"Gemini-3-Pro-Preview[^4]","rowCount":12,"numericRowCount":12,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2026-01-05","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=14;column=All (%)","observedAt":"2026-01-05"}]},{"modelRef":"gpt-4o-2024-08-06","rowCount":12,"numericRowCount":12,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":12},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-10-08","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=33;column=All (%)","observedAt":"2024-10-08"}]},{"modelRef":"dola-seed-2.0-pro","rowCount":11,"numericRowCount":11,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=42","observedAt":"2026-08-21"}]},{"modelRef":"llama-4-scout-17b-16e-instruct","rowCount":11,"numericRowCount":11,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=250","observedAt":"2026-08-21"}]},{"modelRef":"muse-spark-1.2 (xHigh)","rowCount":11,"numericRowCount":11,"benchmarkCount":3,"benchmarkIds":["arena-text","arena-vision","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=419","observedAt":"2026-08-21"}]},{"modelRef":"Kimi-K2-Instruct","rowCount":11,"numericRowCount":11,"benchmarkCount":8,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-gso_external","epoch-lech_mazur_writing_external","epoch-simplebench_external","epoch-terminalbench_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=13;column=Percent correct","observedAt":"2025-07-12"}]},{"modelRef":"Llama-4-Scout-17B-16E-Instruct","rowCount":11,"numericRowCount":11,"benchmarkCount":11,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","epoch-spatialviz_bench_external","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=203;column=Score","observedAt":"2025-04-05"}]},{"modelRef":"Qwen-7B","rowCount":11,"numericRowCount":11,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-lambada_external","epoch-mmlu_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=34;column=Challenge score","observedAt":"2023-09-28"}]},{"modelRef":"claude-sonnet-4-20250514_unknown","rowCount":11,"numericRowCount":11,"benchmarkCount":10,"benchmarkIds":["epoch-apex_agents_external","epoch-critpt_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-gso_external","epoch-mindcube_external","epoch-scicode_external","epoch-simplebench_external","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=50;column=Pass@1 score","observedAt":"2025-05-22"}]},{"modelRef":"gemini-1.5-pro-002","rowCount":11,"numericRowCount":11,"benchmarkCount":11,"benchmarkIds":["epoch-arc_agi_2_external","epoch-balrog_external","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-the_agent_company_external","epoch-weirdml_external","gpqa-diamond","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=194;column=Score","observedAt":"2024-09-24"}]},{"modelRef":"gemini-2.5-pro-preview-05-06","rowCount":11,"numericRowCount":11,"benchmarkCount":10,"benchmarkIds":["aider-polyglot","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-the_agent_company_external","epoch-vpct_external","gpqa-diamond","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-10","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=16;column=Percent correct","observedAt":"2025-05-07"}]},{"modelRef":"gemini-3.1-flash-lite","rowCount":11,"numericRowCount":11,"benchmarkCount":11,"benchmarkIds":["epoch-ale_bench_external","epoch-apex_agents_external","epoch-critpt_external","epoch-deepresearchbench_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-scicode_external","epoch-webdev_arena_external","epoch-weirdml_external","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-03-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=60;column=Performance","observedAt":"2026-03-03"}]},{"modelRef":"gemma-3-27b-it","rowCount":11,"numericRowCount":11,"benchmarkCount":11,"benchmarkIds":["aider-polyglot","epoch-chess_puzzles","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=54;column=Percent correct","observedAt":"2025-03-15"}]},{"modelRef":"gpt-4.1-nano-2025-04-14","rowCount":11,"numericRowCount":11,"benchmarkCount":11,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","epoch-weirdml_external","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=28;column=Percent correct","observedAt":"2025-04-14"}]},{"modelRef":"gpt-4o-2024-08-06","rowCount":11,"numericRowCount":11,"benchmarkCount":11,"benchmarkIds":["aider-polyglot","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","frontiermath","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-30","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=66;column=Percent correct","observedAt":"2024-12-30"}]},{"modelRef":"muse-spark","rowCount":11,"numericRowCount":11,"benchmarkCount":11,"benchmarkIds":["epoch-critpt_external","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-otis_mock_aime_2024_2025","epoch-proofbench_external","epoch-scicode_external","epoch-webdev_arena_external","frontiermath","gpqa-diamond","hle","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-08","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=40;column=Accuracy","observedAt":"2026-04-08"}]},{"modelRef":"o1-2024-12-17_high","rowCount":11,"numericRowCount":11,"benchmarkCount":11,"benchmarkIds":["aider-polyglot","epoch-apex_agents_external","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-weirdml_external","frontiermath","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-21","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=63;column=Percent correct","observedAt":"2024-12-21"}]},{"modelRef":"o1-2024-12-17_medium","rowCount":11,"numericRowCount":11,"benchmarkCount":11,"benchmarkIds":["epoch-arc_agi_external","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-vpct_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arcprize.org/leaderboard","sourceLocator":"arc_agi_external.csv:row=175;column=Score","observedAt":"2024-12-17"}]},{"modelRef":"o3-mini-2025-01-31_medium","rowCount":11,"numericRowCount":11,"benchmarkCount":11,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-31","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=64;column=Percent correct","observedAt":"2025-01-31"}]},{"modelRef":"phi-2","rowCount":11,"numericRowCount":11,"benchmarkCount":9,"benchmarkIds":["epoch-adversarial_nli_external","epoch-arc_ai2_external","epoch-bbh_external","epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-12-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"adversarial_nli_external.csv:row=8;column=Score","observedAt":"2023-12-12"}]},{"modelRef":"qwen2.5-72b-instruct","rowCount":11,"numericRowCount":11,"benchmarkCount":10,"benchmarkIds":["epoch-balrog_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-the_agent_company_external","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":11},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=29;column=Average progress","observedAt":"2024-09-19"}]},{"modelRef":"Llama-3.1-405B-Instruct","rowCount":10,"numericRowCount":10,"benchmarkCount":10,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-the_agent_company_external","epoch-weirdml_external","epoch-wino_grande_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=297;column=ECI Score","observedAt":"2024-07-23"}]},{"modelRef":"Phi-3-small-8k-instruct","rowCount":10,"numericRowCount":10,"benchmarkCount":10,"benchmarkIds":["epoch-adversarial_nli_external","epoch-arc_ai2_external","epoch-bbh_external","epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"adversarial_nli_external.csv:row=6;column=Score","observedAt":"2024-04-23"}]},{"modelRef":"StableBeluga2","rowCount":10,"numericRowCount":10,"benchmarkCount":9,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-07-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=32;column=Challenge score","observedAt":"2023-07-20"}]},{"modelRef":"gemini-1.5-flash-002","rowCount":10,"numericRowCount":10,"benchmarkCount":10,"benchmarkIds":["epoch-balrog_external","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-piqa_external","epoch-weirdml_external","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=31;column=Average progress","observedAt":"2024-09-24"}]},{"modelRef":"gemini-2.0-flash-001","rowCount":10,"numericRowCount":10,"benchmarkCount":10,"benchmarkIds":["epoch-arc_agi_2_external","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-the_agent_company_external","epoch-weirdml_external","frontiermath","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=174;column=Score","observedAt":"2025-02-05"}]},{"modelRef":"gemini-2.5-pro-exp-03-25","rowCount":10,"numericRowCount":10,"benchmarkCount":10,"benchmarkIds":["aider-polyglot","epoch-balrog_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-lech_mazur_writing_external","epoch-simplebench_external","gpqa-diamond","hle","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=36;column=Percent correct","observedAt":"2025-03-25"}]},{"modelRef":"gemma-2b","rowCount":10,"numericRowCount":10,"benchmarkCount":10,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-21","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.08295","sourceLocator":"arc_ai2_external.csv:row=91;column=Challenge score","observedAt":"2024-02-21"}]},{"modelRef":"gpt-4-turbo-2024-04-09","rowCount":10,"numericRowCount":10,"benchmarkCount":10,"benchmarkIds":["epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-09","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=108;column=mean_score","observedAt":"2024-04-09"}]},{"modelRef":"gpt-oss-120b","rowCount":10,"numericRowCount":10,"benchmarkCount":9,"benchmarkIds":["epoch-ale_bench_external","epoch-apex_agents_external","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-metr_time_horizons_external","epoch-simplebench_external","epoch-surface_evolver_bench_external","epoch-terminalbench_external","epoch-vending_bench_2_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=86;column=Performance","observedAt":"2025-08-05"}]},{"modelRef":"internlm-20b","rowCount":10,"numericRowCount":10,"benchmarkCount":7,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=22;column=Challenge score","observedAt":"2023-09-18"}]},{"modelRef":"muse-spark-1.1","rowCount":10,"numericRowCount":10,"benchmarkCount":10,"benchmarkIds":["epoch-apex_agents_external","epoch-critpt_external","epoch-deepswe_external","epoch-epoch_capabilities_index","epoch-gbaeval_external","epoch-gdp_pdf_external","epoch-proofbench_external","epoch-scicode_external","epoch-vending_bench_2_external","epoch-webdev_arena_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-09","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=5;column=Pass@1 score","observedAt":"2026-07-09"}]},{"modelRef":"o1-mini-2024-09-12_medium","rowCount":10,"numericRowCount":10,"benchmarkCount":10,"benchmarkIds":["epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-weirdml_external","frontiermath","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arcprize.org/leaderboard","sourceLocator":"arc_agi_external.csv:row=210;column=Score","observedAt":"2024-09-12"}]},{"modelRef":"openai/gpt-oss-120b_high","rowCount":10,"numericRowCount":10,"benchmarkCount":10,"benchmarkIds":["aider-polyglot","epoch-algotune_external","epoch-chess_puzzles","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","epoch-weirdml_external","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":10},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=76;column=Percent correct","observedAt":"2025-08-06"}]},{"modelRef":"claude-3-5-sonnet-20241022","rowCount":9,"numericRowCount":9,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=229","observedAt":"2026-08-21"}]},{"modelRef":"claude-3-7-sonnet-20250219","rowCount":9,"numericRowCount":9,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=226","observedAt":"2026-08-21"}]},{"modelRef":"claude-sonnet-4-20250514","rowCount":9,"numericRowCount":9,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=196","observedAt":"2026-08-21"}]},{"modelRef":"ernie-5.0-preview-1220","rowCount":9,"numericRowCount":9,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=233","observedAt":"2026-08-21"}]},{"modelRef":"glm-4.6v","rowCount":9,"numericRowCount":9,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=159","observedAt":"2026-08-21"}]},{"modelRef":"hunyuan-vision-1.5-thinking","rowCount":9,"numericRowCount":9,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=129","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-vl-235b-a22b-thinking","rowCount":9,"numericRowCount":9,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=135","observedAt":"2026-08-21"}]},{"modelRef":"Cerebras-GPT-13B","rowCount":9,"numericRowCount":9,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-03-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.03450","sourceLocator":"arc_ai2_external.csv:row=144;column=Challenge score","observedAt":"2023-03-20"}]},{"modelRef":"DeepSeek-V2","rowCount":9,"numericRowCount":9,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-mmlu_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-05-07","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2412.19437","sourceLocator":"arc_ai2_external.csv:row=69;column=Challenge score","observedAt":"2024-05-07"}]},{"modelRef":"Llama-3.1-405B","rowCount":9,"numericRowCount":9,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-mmlu_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-07-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2412.19437","sourceLocator":"arc_ai2_external.csv:row=71;column=Challenge score","observedAt":"2024-07-23"}]},{"modelRef":"Llama-3.1-70B-Instruct","rowCount":9,"numericRowCount":9,"benchmarkCount":8,"benchmarkIds":["epoch-balrog_external","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-the_agent_company_external","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=20;column=Average progress","observedAt":"2024-07-23"}]},{"modelRef":"Qwen-1_8B","rowCount":9,"numericRowCount":9,"benchmarkCount":7,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-gsm8k_external","epoch-lambada_external","epoch-mmlu_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-30","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=33;column=Challenge score","observedAt":"2023-11-30"}]},{"modelRef":"Qwen2.5-72B","rowCount":9,"numericRowCount":9,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-mmlu_external","epoch-piqa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-19","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2412.19437","sourceLocator":"arc_ai2_external.csv:row=70;column=Challenge score","observedAt":"2024-09-19"}]},{"modelRef":"Qwen3-235B-A22B-Thinking-2507","rowCount":9,"numericRowCount":9,"benchmarkCount":8,"benchmarkIds":["epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-otis_mock_aime_2024_2025","epoch-vending_bench_2_external","epoch-weirdml_external","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-07-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=154;column=mean_score","observedAt":"2025-07-25"}]},{"modelRef":"Yi-34B-Chat","rowCount":9,"numericRowCount":9,"benchmarkCount":6,"benchmarkIds":["epoch-bbh_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-math_level_5","epoch-mmlu_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.04652","sourceLocator":"bbh_external.csv:row=82;column=Average","observedAt":"2023-11-22"}]},{"modelRef":"claude-3-7-sonnet-20250219_16K","rowCount":9,"numericRowCount":9,"benchmarkCount":9,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=195;column=Score","observedAt":"2025-02-24"}]},{"modelRef":"claude-3-haiku-20240307","rowCount":9,"numericRowCount":9,"benchmarkCount":9,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-science_qa_external","epoch-weirdml_external","epoch-wino_grande_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-03-07","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=302;column=ECI Score","observedAt":"2024-03-07"}]},{"modelRef":"dolly-v2-12b","rowCount":9,"numericRowCount":9,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-04-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.03450","sourceLocator":"arc_ai2_external.csv:row=143;column=Challenge score","observedAt":"2023-04-11"}]},{"modelRef":"falcon-11b","rowCount":9,"numericRowCount":9,"benchmarkCount":5,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-05-09","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=416;column=ECI Score","observedAt":"2024-05-09"}]},{"modelRef":"gemini-2.0-flash-thinking-exp-01-21","rowCount":9,"numericRowCount":9,"benchmarkCount":9,"benchmarkIds":["aider-polyglot","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","gpqa-diamond","hle","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-21","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=40;column=Percent correct","observedAt":"2025-01-21"}]},{"modelRef":"gpt-3.5-turbo-1106","rowCount":9,"numericRowCount":9,"benchmarkCount":9,"benchmarkIds":["epoch-adversarial_nli_external","epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-mmlu_external","epoch-open_book_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"adversarial_nli_external.csv:row=13;column=Score","observedAt":"2023-11-06"}]},{"modelRef":"grok-2-1212","rowCount":9,"numericRowCount":9,"benchmarkCount":9,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-weirdml_external","frontiermath","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=283;column=ECI Score","observedAt":"2024-12-12"}]},{"modelRef":"grok-3-beta","rowCount":9,"numericRowCount":9,"benchmarkCount":9,"benchmarkIds":["aider-polyglot","epoch-balrog_external","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-10","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=33;column=Percent correct","observedAt":"2025-04-10"}]},{"modelRef":"grok-4-3","rowCount":9,"numericRowCount":9,"benchmarkCount":9,"benchmarkIds":["epoch-ale_bench_external","epoch-blueprint_bench_2_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-gdp_pdf_external","epoch-vending_bench_2_external","epoch-webdev_arena_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=46;column=Performance","observedAt":"2026-04-17"}]},{"modelRef":"grok-4.3_high","rowCount":9,"numericRowCount":9,"benchmarkCount":9,"benchmarkIds":["epoch-chess_puzzles","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-gdp_pdf_external","epoch-otis_mock_aime_2024_2025","epoch-proofbench_external","epoch-scicode_external","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=125;column=mean_score","observedAt":"2026-04-17"}]},{"modelRef":"internlm-7b","rowCount":9,"numericRowCount":9,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external","epoch-lambada_external","epoch-mmlu_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-07-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2309.16609","sourceLocator":"arc_ai2_external.csv:row=21;column=Challenge score","observedAt":"2023-07-05"}]},{"modelRef":"mistral-medium-2604","rowCount":9,"numericRowCount":9,"benchmarkCount":9,"benchmarkIds":["epoch-ale_bench_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-frontiercode_external","epoch-proofbench_external","epoch-scicode_external","epoch-surface_evolver_bench_external","epoch-webdev_arena_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=67;column=Performance","observedAt":"2026-04-28"}]},{"modelRef":"qwen3-max-2025-09-23","rowCount":9,"numericRowCount":9,"benchmarkCount":9,"benchmarkIds":["epoch-ale_bench_external","epoch-chess_puzzles","epoch-cl_bench_external","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-vending_bench_2_external","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=97;column=Performance","observedAt":"2025-09-24"}]},{"modelRef":"xgen-7b-8k-base","rowCount":9,"numericRowCount":9,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":9},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-06-27","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.03450","sourceLocator":"arc_ai2_external.csv:row=134;column=Challenge score","observedAt":"2023-06-27"}]},{"modelRef":"claude-3-7-sonnet-20250219-thinking-32k","rowCount":8,"numericRowCount":8,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=218","observedAt":"2026-08-21"}]},{"modelRef":"glm-4.5v","rowCount":8,"numericRowCount":8,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=201","observedAt":"2026-08-21"}]},{"modelRef":"step-1o-turbo-202506","rowCount":8,"numericRowCount":8,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=199","observedAt":"2026-08-21"}]},{"modelRef":"step-3","rowCount":8,"numericRowCount":8,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=186","observedAt":"2026-08-21"}]},{"modelRef":"Inkling","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-ale_bench_external","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-frontiercode_external","epoch-proofbench_external","epoch-simplebench_external","epoch-webdev_arena_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=45;column=Performance","observedAt":"2026-07-15"}]},{"modelRef":"Llama-3.2-90B-Vision-Instruct","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-balrog_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=21;column=Average progress","observedAt":"2024-09-24"}]},{"modelRef":"Nemotron-4 15B","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-26","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2402.16819","sourceLocator":"arc_ai2_external.csv:row=45;column=Challenge score","observedAt":"2024-02-26"}]},{"modelRef":"PaLM 2-S","rowCount":8,"numericRowCount":8,"benchmarkCount":6,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-open_book_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-05-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2305.10403","sourceLocator":"arc_ai2_external.csv:row=37;column=Challenge score","observedAt":"2023-05-17"}]},{"modelRef":"PaLM 62B","rowCount":8,"numericRowCount":8,"benchmarkCount":7,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-gsm8k_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-04-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2302.13971","sourceLocator":"arc_ai2_external.csv:row=63;column=Challenge score","observedAt":"2022-04-04"}]},{"modelRef":"RedPajama-INCITE-7B-Base","rowCount":8,"numericRowCount":8,"benchmarkCount":7,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-05-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.03450","sourceLocator":"arc_ai2_external.csv:row=139;column=Challenge score","observedAt":"2023-05-04"}]},{"modelRef":"Yi-6B","rowCount":8,"numericRowCount":8,"benchmarkCount":7,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.04652","sourceLocator":"arc_ai2_external.csv:row=2;column=Challenge score","observedAt":"2023-11-02"}]},{"modelRef":"claude-opus-4-1-20250805_16K","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-ale_bench_external","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-vpct_external","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=74;column=Performance","observedAt":"2025-08-05"}]},{"modelRef":"claude-sonnet-4-20250514_16K","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=124;column=Score","observedAt":"2025-05-22"}]},{"modelRef":"gemini-1.5-flash-001","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-gsm8k_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-05-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"bool_q_external.csv:row=126;column=Score","observedAt":"2024-05-23"}]},{"modelRef":"gemini-2.5-flash-preview-05-20","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-otis_mock_aime_2024_2025","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-26","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=20;column=Percent correct","observedAt":"2025-05-26"}]},{"modelRef":"gemma-4-31b-it","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-ale_bench_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-scicode_external","epoch-surface_evolver_bench_external","epoch-webdev_arena_external","epoch-weirdml_external","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=49;column=Performance","observedAt":"2026-04-02"}]},{"modelRef":"glm-4.6","rowCount":8,"numericRowCount":8,"benchmarkCount":7,"benchmarkIds":["epoch-ale_bench_external","epoch-apex_agents_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-scicode_external","epoch-terminalbench_external","epoch-webdev_arena_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-01-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=100;column=Performance","observedAt":"2025-09-30"}]},{"modelRef":"gpt-3.5-turbo-0125","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-01-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=54;column=mean_score","observedAt":"2024-01-25"}]},{"modelRef":"gpt-3.5-turbo-0613","rowCount":8,"numericRowCount":8,"benchmarkCount":7,"benchmarkIds":["epoch-bbh_external","epoch-bool_q_external","epoch-common_sense_qa_2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-06-13","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.10305","sourceLocator":"bbh_external.csv:row=24;column=Average","observedAt":"2023-06-13"}]},{"modelRef":"gpt-4-0314","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-metr_time_horizons_external","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-wino_grande_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-03-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=228;column=ECI Score","observedAt":"2023-03-14"}]},{"modelRef":"gpt-4o-2024-05-13","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-balrog_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-science_qa_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-05-13","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=15;column=Average progress","observedAt":"2024-05-13"}]},{"modelRef":"gpt-5.5-pre-release_xhigh","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond","simpleqa","swebench-verified"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=132;column=mean_score","observedAt":"2026-04-23"}]},{"modelRef":"gpt-j-6b","rowCount":8,"numericRowCount":8,"benchmarkCount":7,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2021-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.03450","sourceLocator":"arc_ai2_external.csv:row=142;column=Challenge score","observedAt":"2021-08-05"}]},{"modelRef":"gpt2-xl","rowCount":8,"numericRowCount":8,"benchmarkCount":7,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-metr_time_horizons_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2019-11-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.05463","sourceLocator":"arc_ai2_external.csv:row=116;column=Challenge score","observedAt":"2019-11-05"}]},{"modelRef":"grok-3-mini-beta_high","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-weirdml_external","frontiermath","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-10","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=31;column=Percent correct","observedAt":"2025-04-10"}]},{"modelRef":"grok-3-mini-beta_low","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["aider-polyglot","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-10","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=11;column=Percent correct","observedAt":"2025-04-10"}]},{"modelRef":"inkling-small_xhigh","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-mystery_game_puzzles","epoch-otis_mock_aime_2024_2025","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=80;column=Score","observedAt":"2026-07-15"}]},{"modelRef":"kimi-k2-thinking","rowCount":8,"numericRowCount":8,"benchmarkCount":7,"benchmarkIds":["epoch-ale_bench_external","epoch-apex_agents_external","epoch-cl_bench_external","epoch-epoch_capabilities_index","epoch-metr_time_horizons_external","epoch-terminalbench_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=84;column=Performance","observedAt":"2025-11-06"}]},{"modelRef":"mistral-large-2407","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-07-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=305;column=ECI Score","observedAt":"2024-07-24"}]},{"modelRef":"o1-preview-2024-09-12","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-otis_mock_aime_2024_2025","epoch-simplebench_external","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arcprize.org/leaderboard","sourceLocator":"arc_agi_external.csv:row=199;column=Score","observedAt":"2024-09-12"}]},{"modelRef":"o3-mini-2025-01-31_low","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-gso_external","epoch-otis_mock_aime_2024_2025","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-31","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=205;column=Score","observedAt":"2025-01-31"}]},{"modelRef":"o4-mini-2025-04-16_low","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=169;column=Score","observedAt":"2025-04-16"}]},{"modelRef":"open_llama_7b","rowCount":8,"numericRowCount":8,"benchmarkCount":7,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-06-07","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.03450","sourceLocator":"arc_ai2_external.csv:row=138;column=Challenge score","observedAt":"2023-06-07"}]},{"modelRef":"opt-13b","rowCount":8,"numericRowCount":8,"benchmarkCount":7,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-05-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.03450","sourceLocator":"arc_ai2_external.csv:row=141;column=Challenge score","observedAt":"2022-05-11"}]},{"modelRef":"phi-4","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-balrog_external","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=33;column=Average progress","observedAt":"2024-12-12"}]},{"modelRef":"text-davinci-003","rowCount":8,"numericRowCount":8,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-11-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=28;column=Score","observedAt":"2022-11-28"}]},{"modelRef":"vicuna-13b-v1.1","rowCount":8,"numericRowCount":8,"benchmarkCount":8,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-04-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.05463","sourceLocator":"arc_ai2_external.csv:row=108;column=Challenge score","observedAt":"2023-04-12"}]},{"modelRef":"deepseek-r1","rowCount":8,"numericRowCount":8,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-09-12","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=25;column=All (%)","observedAt":"2025-09-12"}]},{"modelRef":"o1-preview","rowCount":8,"numericRowCount":8,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-05-14","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=31;column=All (%)","observedAt":"2025-05-14"}]},{"modelRef":"Claude 3.7 Sonnet","rowCount":8,"numericRowCount":8,"benchmarkCount":5,"benchmarkIds":["swebench-bash-only","swebench-lite","swebench-multimodal","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":8},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-07-20","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[34]","observedAt":"2025-07-20"}]},{"modelRef":"openai/gpt-oss-120b","rowCount":7,"numericRowCount":7,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://arxiv.org/abs/2508.10925","sourceLocator":"rank=61;row=60"}]},{"modelRef":"openai/gpt-oss-20b","rowCount":7,"numericRowCount":7,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://arxiv.org/abs/2508.10925","sourceLocator":"rank=101;row=100"}]},{"modelRef":"claude-3-5-haiku-20241022","rowCount":7,"numericRowCount":7,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=270","observedAt":"2026-08-21"}]},{"modelRef":"claude-sonnet-4-20250514-thinking-32k","rowCount":7,"numericRowCount":7,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=188","observedAt":"2026-08-21"}]},{"modelRef":"granite-4.1-8b","rowCount":7,"numericRowCount":7,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=233","observedAt":"2026-08-21"}]},{"modelRef":"hunyuan-hy3-preview","rowCount":7,"numericRowCount":7,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=132","observedAt":"2026-08-21"}]},{"modelRef":"hunyuan-large-vision","rowCount":7,"numericRowCount":7,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=264","observedAt":"2026-08-21"}]},{"modelRef":"hy3","rowCount":7,"numericRowCount":7,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=425","observedAt":"2026-08-21"}]},{"modelRef":"inkling","rowCount":7,"numericRowCount":7,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=441","observedAt":"2026-08-21"}]},{"modelRef":"kimi-k2-thinking-turbo","rowCount":7,"numericRowCount":7,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=116","observedAt":"2026-08-21"}]},{"modelRef":"muse-glimmer","rowCount":7,"numericRowCount":7,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=146","observedAt":"2026-08-21"}]},{"modelRef":"trinity-large-thinking","rowCount":7,"numericRowCount":7,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=191","observedAt":"2026-08-21"}]},{"modelRef":"INTELLECT-1-Instruct","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-arc_ai2_external","epoch-bbh_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-11-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/abs/2412.01152","sourceLocator":"arc_ai2_external.csv:row=146;column=Challenge score","observedAt":"2024-11-29"}]},{"modelRef":"Inkling_xhigh","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-chess_puzzles","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=83;column=mean_score","observedAt":"2026-07-15"}]},{"modelRef":"Llama-2-13b-chat","rowCount":7,"numericRowCount":7,"benchmarkCount":4,"benchmarkIds":["epoch-bbh_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-07-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.04652","sourceLocator":"bbh_external.csv:row=75;column=Average","observedAt":"2023-07-18"}]},{"modelRef":"Llama-2-70b-chat","rowCount":7,"numericRowCount":7,"benchmarkCount":4,"benchmarkIds":["epoch-bbh_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-07-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.04652","sourceLocator":"bbh_external.csv:row=76;column=Average","observedAt":"2023-07-18"}]},{"modelRef":"Llama-4-Maverick-17B-128E-Instruct-FP8","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-ale_bench_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=110;column=Performance","observedAt":"2025-04-05"}]},{"modelRef":"PaLM 2-M","rowCount":7,"numericRowCount":7,"benchmarkCount":6,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-open_book_qa_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-05-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2305.10403","sourceLocator":"arc_ai2_external.csv:row=38;column=Challenge score","observedAt":"2023-05-17"}]},{"modelRef":"Qwen-14B-Chat","rowCount":7,"numericRowCount":7,"benchmarkCount":4,"benchmarkIds":["epoch-bbh_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.04652","sourceLocator":"bbh_external.csv:row=78;column=Average","observedAt":"2023-09-24"}]},{"modelRef":"Qwen2.5-Coder-32B","rowCount":7,"numericRowCount":7,"benchmarkCount":6,"benchmarkIds":["epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=107;column=Challenge score","observedAt":"2024-09-18"}]},{"modelRef":"Yi-6B-Chat","rowCount":7,"numericRowCount":7,"benchmarkCount":4,"benchmarkIds":["epoch-bbh_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.04652","sourceLocator":"bbh_external.csv:row=81;column=Average","observedAt":"2023-11-22"}]},{"modelRef":"claude-3-7-sonnet-20250219_unknown","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-gso_external","epoch-simplebench_external","epoch-spatialviz_bench_external","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=24;column=Accuracy","observedAt":"2025-02-24"}]},{"modelRef":"claude-3-sonnet-20240229","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-weirdml_external","epoch-wino_grande_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=294;column=ECI Score","observedAt":"2024-02-29"}]},{"modelRef":"claude-instant-1.1","rowCount":7,"numericRowCount":7,"benchmarkCount":5,"benchmarkIds":["epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-trivia_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://www.anthropic.com/news/releasing-claude-instant-1-2","sourceLocator":"arc_ai2_external.csv:row=130;column=Challenge score"}]},{"modelRef":"claude-sonnet-4-20250514_32K","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["aider-polyglot","epoch-ale_bench_external","epoch-epoch_capabilities_index","epoch-geobench_external","epoch-otis_mock_aime_2024_2025","epoch-vpct_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=22;column=Percent correct","observedAt":"2025-05-24"}]},{"modelRef":"gemini-1.5-pro-001","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-bbh_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-05-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://storage.googleapis.com/deepmind-media/gemini/gemini_v1_5_report.pdf","sourceLocator":"bbh_external.csv:row=93;column=Average","observedAt":"2024-05-14"}]},{"modelRef":"gemini-2.5-flash-preview-04-17","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-geobench_external","epoch-otis_mock_aime_2024_2025","epoch-vpct_external","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=30;column=Percent correct","observedAt":"2025-04-20"}]},{"modelRef":"gemini-2.5-flash-preview-09-2025","rowCount":7,"numericRowCount":7,"benchmarkCount":4,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-terminalbench_external","epoch-the_agent_company_external","epoch-vpct_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=341;column=ECI Score","observedAt":"2025-09-25"}]},{"modelRef":"gemini-2.5-pro-preview-03-25","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["aider-polyglot","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-simplebench_external","epoch-vpct_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=71;column=Percent correct","observedAt":"2025-04-12"}]},{"modelRef":"gpt-5-codex","rowCount":7,"numericRowCount":7,"benchmarkCount":2,"benchmarkIds":["epoch-terminalbench_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0","sourceLocator":"terminalbench_external.csv:row=104;column=Accuracy mean","observedAt":"2025-11-03"}]},{"modelRef":"gpt-5.5-pro-pre-release_xhigh","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=131;column=mean_score","observedAt":"2026-04-23"}]},{"modelRef":"grok-4.20-0309-reasoning","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-otis_mock_aime_2024_2025","epoch-proofbench_external","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-02-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=115;column=mean_score","observedAt":"2026-02-17"}]},{"modelRef":"grok-code-fast-1","rowCount":7,"numericRowCount":7,"benchmarkCount":4,"benchmarkIds":["epoch-ale_bench_external","epoch-terminalbench_external","epoch-webdev_arena_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-01-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=85;column=Performance","observedAt":"2025-08-28"}]},{"modelRef":"mistral-large-2411","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-11-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=284;column=ECI Score","observedAt":"2024-11-18"}]},{"modelRef":"mistral-medium-2505","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-07","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=251;column=ECI Score","observedAt":"2025-05-07"}]},{"modelRef":"mistral-small-2503","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-critpt_external","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=117;column=Accuracy","observedAt":"2025-03-17"}]},{"modelRef":"o3-2025-04-16_low","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=161;column=Score","observedAt":"2025-04-16"}]},{"modelRef":"phi-1_5","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-mmlu_external","epoch-open_book_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.05463","sourceLocator":"arc_ai2_external.csv:row=119;column=Challenge score","observedAt":"2023-09-11"}]},{"modelRef":"qwen2-72b-instruct","rowCount":7,"numericRowCount":7,"benchmarkCount":7,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-metr_time_horizons_external","epoch-mmlu_external","epoch-the_agent_company_external","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=312;column=ECI Score","observedAt":"2024-06-07"}]},{"modelRef":"GPT-4 (1106)","rowCount":7,"numericRowCount":7,"benchmarkCount":3,"benchmarkIds":["swebench-lite","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":7},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-06-04","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[54]","observedAt":"2024-06-04"}]},{"modelRef":"openai/gpt-oss-120b","rowCount":6,"numericRowCount":6,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://arxiv.org/abs/2508.10925","sourceLocator":"rank=65;row=64"}]},{"modelRef":"openai/gpt-oss-20b","rowCount":6,"numericRowCount":6,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://arxiv.org/abs/2508.10925","sourceLocator":"rank=71;row=70"}]},{"modelRef":"claude-3-5-sonnet-20240620","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=248","observedAt":"2026-08-21"}]},{"modelRef":"claude-opus-4-1-20250805","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=108","observedAt":"2026-08-21"}]},{"modelRef":"gemini-1.5-pro-002","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=214","observedAt":"2026-08-21"}]},{"modelRef":"gemini-2.0-flash-lite-preview-02-05","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=206","observedAt":"2026-08-21"}]},{"modelRef":"glm-4.6","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=437","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4.5-preview-2025-02-27","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=109","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4o-2024-05-13","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=225","observedAt":"2026-08-21"}]},{"modelRef":"grok-4.1-thinking","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=501","observedAt":"2026-08-21"}]},{"modelRef":"mercury-2","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=179","observedAt":"2026-08-21"}]},{"modelRef":"minimax-m2","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=189","observedAt":"2026-08-21"}]},{"modelRef":"o1-2024-12-17","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=170","observedAt":"2026-08-21"}]},{"modelRef":"solar-pro4","rowCount":6,"numericRowCount":6,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=156","observedAt":"2026-08-21"}]},{"modelRef":"Baichuan2-13B-Chat","rowCount":6,"numericRowCount":6,"benchmarkCount":3,"benchmarkIds":["epoch-bbh_external","epoch-gsm8k_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.04652","sourceLocator":"bbh_external.csv:row=77;column=Average","observedAt":"2023-09-06"}]},{"modelRef":"PaLM 2-L","rowCount":6,"numericRowCount":6,"benchmarkCount":5,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-trivia_qa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-05-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2305.10403","sourceLocator":"arc_ai2_external.csv:row=39;column=Challenge score","observedAt":"2023-05-17"}]},{"modelRef":"Qwen2.5-Coder-1.5B","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=95;column=Challenge score","observedAt":"2024-09-18"}]},{"modelRef":"Qwen2.5-Coder-7B","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=102;column=Challenge score","observedAt":"2024-09-18"}]},{"modelRef":"amazon.nova-pro-v1:0","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-mmlu_external","epoch-the_agent_company_external","hle","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=521;column=ECI Score","observedAt":"2024-12-03"}]},{"modelRef":"claude-2.0","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-trivia_qa_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-07-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=275;column=ECI Score","observedAt":"2023-07-11"}]},{"modelRef":"claude-2.1","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-21","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=274;column=ECI Score","observedAt":"2023-11-21"}]},{"modelRef":"claude-3-7-sonnet-20250219_32K","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=68;column=Percent correct","observedAt":"2025-02-24"}]},{"modelRef":"claude-3-7-sonnet-20250219_64K","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","epoch-vpct_external","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=239;column=ECI Score","observedAt":"2025-02-24"}]},{"modelRef":"claude-instant-1.2","rowCount":6,"numericRowCount":6,"benchmarkCount":5,"benchmarkIds":["epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-trivia_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-08-09","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://www.anthropic.com/news/releasing-claude-instant-1-2","sourceLocator":"arc_ai2_external.csv:row=131;column=Challenge score","observedAt":"2023-08-09"}]},{"modelRef":"claude-opus-4-1-20250805_27K","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=219;column=ECI Score","observedAt":"2025-08-05"}]},{"modelRef":"deepseek-reasoner","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["aider-polyglot","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-12-01","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=77;column=Percent correct","observedAt":"2025-10-03"}]},{"modelRef":"gemini-2.0-flash-exp","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-mmlu_external","epoch-simplebench_external","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=58;column=Percent correct","observedAt":"2024-12-22"}]},{"modelRef":"gemini-2.0-pro-exp-02-05","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["aider-polyglot","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-math_level_5","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=49;column=Percent correct","observedAt":"2025-02-25"}]},{"modelRef":"gemini-3-deep-think-preview","rowCount":6,"numericRowCount":6,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-critpt_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-02-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=15;column=Score","observedAt":"2026-02-12"}]},{"modelRef":"gemini-3-flash-preview_high","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-chess_puzzles","epoch-deepresearchbench_external","epoch-epoch_capabilities_index","epoch-mystery_game_puzzles","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=71;column=mean_score","observedAt":"2025-12-17"}]},{"modelRef":"gemma-2-27b-it","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-06-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=272;column=ECI Score","observedAt":"2024-06-24"}]},{"modelRef":"gemma-2-9b-it","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-06-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=273;column=ECI Score","observedAt":"2024-06-24"}]},{"modelRef":"gpt-5-pro-2025-10-06_unknown","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-simplebench_external","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-10-07","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=100;column=Score","observedAt":"2025-10-07"}]},{"modelRef":"gpt-neox-20b","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-04-07","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.03450","sourceLocator":"arc_ai2_external.csv:row=140;column=Challenge score","observedAt":"2022-04-07"}]},{"modelRef":"gpt-oss-20b_high","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-critpt_external","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","epoch-scicode_external","epoch-weirdml_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=79;column=Accuracy","observedAt":"2025-08-05"}]},{"modelRef":"grok-3","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-apex_agents_external","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-simplebench_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-09","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=60;column=Pass@1 score","observedAt":"2025-04-09"}]},{"modelRef":"grok-4-1-fast-reasoning","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-ale_bench_external","epoch-forecastbench_external","epoch-proofbench_external","epoch-simplebench_external","epoch-vending_bench_2_external","epoch-webdev_arena_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-01-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=96;column=Performance","observedAt":"2025-11-19"}]},{"modelRef":"kimi-k2-thinking-turbo","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","epoch-webdev_arena_external","gpqa-diamond","simpleqa"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-01-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=156;column=mean_score","observedAt":"2025-11-06"}]},{"modelRef":"nemotron-3-ultra","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-apex_agents_external","epoch-critpt_external","epoch-epoch_capabilities_index","epoch-proofbench_external","epoch-scicode_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-06-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=49;column=Pass@1 score","observedAt":"2026-06-04"}]},{"modelRef":"opt-1.3b","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-05-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.05463","sourceLocator":"arc_ai2_external.csv:row=114;column=Challenge score","observedAt":"2022-05-11"}]},{"modelRef":"qwen-max-2025-01-25","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=48;column=Percent correct","observedAt":"2025-01-28"}]},{"modelRef":"stablelm-tuned-alpha-7b","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-hella_swag_external","epoch-open_book_qa_external","epoch-piqa_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-04-19","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.03450","sourceLocator":"arc_ai2_external.csv:row=145;column=Challenge score","observedAt":"2023-04-19"}]},{"modelRef":"text-davinci-002","rowCount":6,"numericRowCount":6,"benchmarkCount":6,"benchmarkIds":["epoch-arc_ai2_external","epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-03-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2303.08774","sourceLocator":"arc_ai2_external.csv:row=121;column=Challenge score","observedAt":"2022-03-15"}]},{"modelRef":"Claude 3 Opus","rowCount":6,"numericRowCount":6,"benchmarkCount":3,"benchmarkIds":["swebench-lite","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":6},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-04-02","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[77]","observedAt":"2024-04-02"}]},{"modelRef":"claude-3-haiku-20240307","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=298","observedAt":"2026-08-21"}]},{"modelRef":"claude-3-opus-20240229","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=266","observedAt":"2026-08-21"}]},{"modelRef":"claude-3-sonnet-20240229","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=288","observedAt":"2026-08-21"}]},{"modelRef":"gemini-1.5-flash-001","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=274","observedAt":"2026-08-21"}]},{"modelRef":"gemini-1.5-flash-002","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=239","observedAt":"2026-08-21"}]},{"modelRef":"gemini-1.5-flash-8b-001","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=284","observedAt":"2026-08-21"}]},{"modelRef":"gemini-1.5-pro-001","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=255","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4-turbo-2024-04-09","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=256","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4.1-nano-2025-04-14","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=241","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4o-2024-08-06","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=244","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4o-mini-2024-07-18","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=240","observedAt":"2026-08-21"}]},{"modelRef":"molmo-2-8b","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=230","observedAt":"2026-08-21"}]},{"modelRef":"qwen-vl-max-2025-08-13","rowCount":5,"numericRowCount":5,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=312","observedAt":"2026-01-09"}]},{"modelRef":"seed-2.1-pro-preview","rowCount":5,"numericRowCount":5,"benchmarkCount":1,"benchmarkIds":["arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-25","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=138","observedAt":"2026-08-25"}]},{"modelRef":"Baichuan-7B","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-bbh_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-06-01","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.10305","sourceLocator":"bbh_external.csv:row=30;column=Average","observedAt":"2023-06-01"}]},{"modelRef":"Llama-2-70b-chat-hf","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-07-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=290;column=ECI Score","observedAt":"2023-07-18"}]},{"modelRef":"Meta-Llama-3-70B-Instruct","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=286;column=ECI Score","observedAt":"2024-04-18"}]},{"modelRef":"Qwen2.5-Coder-0.5B","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-arc_ai2_external","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=93;column=Challenge score","observedAt":"2024-09-18"}]},{"modelRef":"Qwen2.5-Coder-14B","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-arc_ai2_external","epoch-gsm8k_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=104;column=Challenge score","observedAt":"2024-09-18"}]},{"modelRef":"Qwen2.5-VL-72B-Instruct","rowCount":5,"numericRowCount":5,"benchmarkCount":4,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-geobench_external","epoch-spatialviz_bench_external","osworld"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-19","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=613;column=ECI Score","observedAt":"2024-09-19"}]},{"modelRef":"Qwen3-235B-A22B-Instruct-2507","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-07-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=72;column=Percent correct","observedAt":"2025-05-09"}]},{"modelRef":"Yi-34B","rowCount":5,"numericRowCount":5,"benchmarkCount":4,"benchmarkIds":["epoch-bbh_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.04652","sourceLocator":"bbh_external.csv:row=15;column=Average","observedAt":"2023-11-02"}]},{"modelRef":"deepseek-coder-1.3b-base","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=94;column=Challenge score","observedAt":"2023-11-02"}]},{"modelRef":"deepseek-coder-33b-base","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=105;column=Challenge score","observedAt":"2023-11-02"}]},{"modelRef":"deepseek-coder-6.7b-base","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=99;column=Challenge score","observedAt":"2023-11-02"}]},{"modelRef":"gemini-1.0-pro-001","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-12-13","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=295;column=ECI Score","observedAt":"2023-12-13"}]},{"modelRef":"gemini-3.1-flash-lite_low","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-chess_puzzles","epoch-deepresearchbench_external","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-03-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=69;column=mean_score","observedAt":"2026-03-03"}]},{"modelRef":"gemma-4-26b-a4b","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-ale_bench_external","epoch-critpt_external","epoch-scicode_external","epoch-webdev_arena_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=48;column=Performance","observedAt":"2026-04-02"}]},{"modelRef":"gpt-5-pro-2025-10-06_high","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","epoch-simplebench_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-10-07","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arcprize.org/leaderboard","sourceLocator":"arc_agi_external.csv:row=108;column=Score","observedAt":"2025-10-07"}]},{"modelRef":"inkling-small","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-critpt_external","epoch-epoch_capabilities_index","epoch-proofbench_external","epoch-scicode_external","epoch-webdev_arena_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=49;column=Accuracy","observedAt":"2026-07-15"}]},{"modelRef":"internlm-chat-20b","rowCount":5,"numericRowCount":5,"benchmarkCount":3,"benchmarkIds":["epoch-bbh_external","epoch-gsm8k_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.04652","sourceLocator":"bbh_external.csv:row=79;column=Average","observedAt":"2023-09-17"}]},{"modelRef":"magistral-small-2506","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-10","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=209;column=Score","observedAt":"2025-06-10"}]},{"modelRef":"mercury-2","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-ale_bench_external","epoch-critpt_external","epoch-scicode_external","epoch-webdev_arena_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=62;column=Performance"}]},{"modelRef":"mistral-large-2402","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-mmlu_external","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-26","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=293;column=ECI Score","observedAt":"2024-02-26"}]},{"modelRef":"muse-spark-1.2_xhigh","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-deepswe_external","epoch-epoch_capabilities_index","epoch-gdp_pdf_external","epoch-webdev_arena_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepswe_external.csv:row=62;column=Pass@1","observedAt":"2026-08-05"}]},{"modelRef":"o1-2024-12-17_low","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-arc_agi_external","epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arcprize.org/leaderboard","sourceLocator":"arc_agi_external.csv:row=183;column=Score","observedAt":"2024-12-17"}]},{"modelRef":"o1-mini-2024-09-12_high","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-otis_mock_aime_2024_2025","frontiermath","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=280;column=ECI Score","observedAt":"2024-09-12"}]},{"modelRef":"o3-pro-2025-06-10_high","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=15;column=Percent correct","observedAt":"2025-06-28"}]},{"modelRef":"qwen3-235b-a22b","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","epoch-math_level_5","epoch-simplebench_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=241;column=ECI Score","observedAt":"2025-04-29"}]},{"modelRef":"starcoder2-15b","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=103;column=Challenge score","observedAt":"2024-02-20"}]},{"modelRef":"starcoder2-3b","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=96;column=Challenge score","observedAt":"2024-02-22"}]},{"modelRef":"starcoder2-7b","rowCount":5,"numericRowCount":5,"benchmarkCount":5,"benchmarkIds":["epoch-arc_ai2_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=98;column=Challenge score","observedAt":"2024-02-20"}]},{"modelRef":"Qwen3-Coder-30B-A3B-Instruct","rowCount":5,"numericRowCount":5,"benchmarkCount":2,"benchmarkIds":["swebench-lite","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":5},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-09-01","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[17]","observedAt":"2025-09-01"}]},{"modelRef":"amazon-nova-lite-v1.0","rowCount":4,"numericRowCount":4,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=280","observedAt":"2026-08-21"}]},{"modelRef":"amazon-nova-pro-v1.0","rowCount":4,"numericRowCount":4,"benchmarkCount":2,"benchmarkIds":["arena-text","arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=268","observedAt":"2026-08-21"}]},{"modelRef":"devstral-2","rowCount":4,"numericRowCount":4,"benchmarkCount":1,"benchmarkIds":["arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=112","observedAt":"2026-08-21"}]},{"modelRef":"ernie-5.1","rowCount":4,"numericRowCount":4,"benchmarkCount":2,"benchmarkIds":["arena-search","arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=5","observedAt":"2026-08-24"}]},{"modelRef":"grok-4.20-beta1","rowCount":4,"numericRowCount":4,"benchmarkCount":2,"benchmarkIds":["arena-search","arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=17","observedAt":"2026-08-24"}]},{"modelRef":"laguna-m.1","rowCount":4,"numericRowCount":4,"benchmarkCount":1,"benchmarkIds":["arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=249","observedAt":"2026-08-21"}]},{"modelRef":"laguna-xs.2","rowCount":4,"numericRowCount":4,"benchmarkCount":1,"benchmarkIds":["arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=258","observedAt":"2026-08-21"}]},{"modelRef":"CodeQwen1.5-7B","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-arc_ai2_external","epoch-gsm8k_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=101;column=Challenge score","observedAt":"2024-04-15"}]},{"modelRef":"DeepSeek-Coder-V2-Lite-Base","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-arc_ai2_external","epoch-gsm8k_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-06-13","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=100;column=Challenge score","observedAt":"2024-06-13"}]},{"modelRef":"DeepSeek-R1-Distill-Llama-70B","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-math_level_5","epoch-otis_mock_aime_2024_2025","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=39;column=mean_score","observedAt":"2025-01-20"}]},{"modelRef":"Inflection-1","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-bool_q_external","epoch-lambada_external","epoch-mmlu_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-06-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2311.16867","sourceLocator":"bool_q_external.csv:row=183;column=Score","observedAt":"2023-06-22"}]},{"modelRef":"Mistral-7B-Instruct-v0.2","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-mmlu_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-12-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.08295","sourceLocator":"bool_q_external.csv:row=23;column=Score","observedAt":"2023-12-11"}]},{"modelRef":"Mistral-Nemo-Base-2407","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-07-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"bool_q_external.csv:row=123;column=Score","observedAt":"2024-07-18"}]},{"modelRef":"Mixtral-8x7B-Instruct-v0.1","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-12-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=317;column=ECI Score","observedAt":"2023-12-11"}]},{"modelRef":"Qwen2.5-Coder-32B-Instruct","rowCount":4,"numericRowCount":4,"benchmarkCount":3,"benchmarkIds":["aider-polyglot","epoch-gsm8k_external","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-11-21","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=43;column=Percent correct","observedAt":"2024-11-21"}]},{"modelRef":"Qwen2.5-Coder-3B","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-arc_ai2_external","epoch-gsm8k_external","epoch-hella_swag_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=97;column=Challenge score","observedAt":"2024-09-18"}]},{"modelRef":"Qwen3-235B-A22B","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/","sourceLocator":"aider_polyglot_external.csv:row=3;column=Percent correct","observedAt":"2025-04-29"}]},{"modelRef":"T5-11B","rowCount":4,"numericRowCount":4,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-common_sense_qa_2_external","epoch-superglue_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/pdf/1910.10683","sourceLocator":"bool_q_external.csv:row=207;column=Score"}]},{"modelRef":"T5-Large","rowCount":4,"numericRowCount":4,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-common_sense_qa_2_external","epoch-superglue_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/pdf/1910.10683","sourceLocator":"bool_q_external.csv:row=205;column=Score"}]},{"modelRef":"Yi-9B","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-arc_ai2_external","epoch-hella_swag_external","epoch-mmlu_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-03-01","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2403.04652","sourceLocator":"arc_ai2_external.csv:row=3;column=Challenge score","observedAt":"2024-03-01"}]},{"modelRef":"amazon.nova-lite-v1:0","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-ale_bench_external","epoch-mmlu_external","hle","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=106;column=Performance","observedAt":"2024-12-03"}]},{"modelRef":"gemini-2.5-flash-preview-05-20_16K","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=162;column=Score","observedAt":"2025-05-20"}]},{"modelRef":"gemini-2.5-flash-preview-05-20_23K","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=19;column=Percent correct","observedAt":"2025-05-25"}]},{"modelRef":"gemini-2.5-pro-preview-06-05_1K","rowCount":4,"numericRowCount":4,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=211;column=Score","observedAt":"2025-06-05"}]},{"modelRef":"gemini-2.5-pro_16K","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=139;column=Score","observedAt":"2025-06-17"}]},{"modelRef":"gemini-3.1-flash-lite_high","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-03-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=70;column=mean_score","observedAt":"2026-03-03"}]},{"modelRef":"gemini-3.1-flash-lite_minimal","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-03-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=68;column=mean_score","observedAt":"2026-03-03"}]},{"modelRef":"gemini-exp-1206","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index","epoch-simplebench_external","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=45;column=Percent correct","observedAt":"2024-12-22"}]},{"modelRef":"gemma-2-9b","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"bool_q_external.csv:row=125;column=Score"}]},{"modelRef":"gemma-4-31b-it_minimal","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-chess_puzzles","epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"chess_puzzles.csv:row=59;column=mean_score","observedAt":"2026-04-02"}]},{"modelRef":"gpt-4-0125-preview","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-metr_time_horizons_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-01-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=307;column=ECI Score","observedAt":"2024-01-25"}]},{"modelRef":"gpt-4-1106-preview","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","epoch-metr_time_horizons_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=309;column=ECI Score","observedAt":"2023-11-06"}]},{"modelRef":"gpt-neo-2.7B","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-bool_q_external","epoch-hella_swag_external","epoch-open_book_qa_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-03-30","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.05463","sourceLocator":"bool_q_external.csv:row=164;column=Score","observedAt":"2023-03-30"}]},{"modelRef":"gpt-oss-20b","rowCount":4,"numericRowCount":4,"benchmarkCount":3,"benchmarkIds":["epoch-ale_bench_external","epoch-epoch_capabilities_index","epoch-terminalbench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=87;column=Performance","observedAt":"2025-08-05"}]},{"modelRef":"magistral-medium-2506","rowCount":4,"numericRowCount":4,"benchmarkCount":2,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-10","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=210;column=Score","observedAt":"2025-06-10"}]},{"modelRef":"mistral-large-2512","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-ale_bench_external","epoch-critpt_external","epoch-scicode_external","epoch-webdev_arena_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-12-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=105;column=Performance","observedAt":"2025-12-02"}]},{"modelRef":"mistral-medium-2508","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-ale_bench_external","epoch-critpt_external","epoch-scicode_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=109;column=Performance","observedAt":"2025-08-12"}]},{"modelRef":"mistral-small-2501","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-math_level_5","epoch-otis_mock_aime_2024_2025","gpqa-diamond","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=57;column=mean_score","observedAt":"2025-01-25"}]},{"modelRef":"moonshotai/Kimi-K2-Thinking","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-cl_bench_external","epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","frontiermath"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cl_bench_external.csv:row=16;column=Overall","observedAt":"2025-11-06"}]},{"modelRef":"muse-spark-1.2","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-gdp_pdf_external","epoch-proofbench_external","epoch-simplebench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=534;column=ECI Score","observedAt":"2026-08-05"}]},{"modelRef":"nemotron-3-super","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-ale_bench_external","epoch-critpt_external","epoch-scicode_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-03-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=108;column=Performance","observedAt":"2026-03-11"}]},{"modelRef":"o1-2024-12-17_unknown","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-enigma_eval_external","epoch-epoch_capabilities_index","epoch-spatialviz_bench_external","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=21;column=Accuracy","observedAt":"2024-12-17"}]},{"modelRef":"o1-mini-2024-09-12_unknown","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["aider-polyglot","epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=70;column=Percent correct","observedAt":"2024-12-22"}]},{"modelRef":"o3-2025-04-16_unknown","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-metr_time_horizons_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=73;column=Percent correct","observedAt":"2025-06-25"}]},{"modelRef":"o3-pro-2025-06-10_medium","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-10","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=163;column=Score","observedAt":"2025-06-10"}]},{"modelRef":"qwen/qwen3-235b-a22b-thinking-2507","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-critpt_external","epoch-epoch_capabilities_index","epoch-scicode_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-07-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=134;column=Accuracy","observedAt":"2025-07-25"}]},{"modelRef":"trinity-large-thinking","rowCount":4,"numericRowCount":4,"benchmarkCount":4,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external","epoch-surface_evolver_bench_external","epoch-webdev_arena_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-01","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=90;column=Accuracy","observedAt":"2026-04-01"}]},{"modelRef":"Deepseek-V3.2-Speciale","rowCount":4,"numericRowCount":4,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-12-16","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=16;column=All (%)","observedAt":"2025-12-16"}]},{"modelRef":"claude-3-5-sonnet-20240620","rowCount":4,"numericRowCount":4,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-10-08","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=34;column=All (%)","observedAt":"2024-10-08"}]},{"modelRef":"gpt-5-codex","rowCount":4,"numericRowCount":4,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-11-10","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=20;column=All (%)","observedAt":"2025-11-10"}]},{"modelRef":"llama-3.1-405b-instruct","rowCount":4,"numericRowCount":4,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-10-08","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=36;column=All (%)","observedAt":"2024-10-08"}]},{"modelRef":"o3","rowCount":4,"numericRowCount":4,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-05-15","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=28;column=All (%)","observedAt":"2025-05-15"}]},{"modelRef":"o3 + GPT-4.1","rowCount":4,"numericRowCount":4,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-08-15","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=29;column=All (%)","observedAt":"2025-08-15"}]},{"modelRef":"undisclosed","rowCount":4,"numericRowCount":4,"benchmarkCount":1,"benchmarkIds":["mle-bench"],"sourceIds":["src-mle-bench"],"sourceLabels":["OpenAI · MLE-bench"],"sourceUrls":["https://raw.githubusercontent.com/openai/mle-bench/main/README.md"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-07-28","examples":[{"sourceId":"src-mle-bench","sourceLabel":"OpenAI · MLE-bench","sourceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","evidenceUrl":"https://raw.githubusercontent.com/openai/mle-bench/main/README.md","sourceLocator":"README.md:row=27;column=All (%)","observedAt":"2025-07-28"}]},{"modelRef":"Qwen2.5 (7B + 72B)","rowCount":4,"numericRowCount":4,"benchmarkCount":2,"benchmarkIds":["swebench-lite","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-03-06","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[64]","observedAt":"2025-03-06"}]},{"modelRef":"o3","rowCount":4,"numericRowCount":4,"benchmarkCount":3,"benchmarkIds":["swebench-bash-only","swebench-multimodal","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-11-17","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[26]","observedAt":"2025-07-26"}]},{"modelRef":"o4-mini","rowCount":4,"numericRowCount":4,"benchmarkCount":3,"benchmarkIds":["swebench-bash-only","swebench-multimodal","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":4},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-07-26","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[35]","observedAt":"2025-07-26"}]},{"modelRef":"RedHatAI/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=55;row=54"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-MXFP4_MOE-dequant-bf16-vllm","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=58;row=57"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q3_K_M-dequant-bf16-vllm","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=56;row=55"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_M-dequant-bf16-vllm","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=57;row=56"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=54;row=53"}]},{"modelRef":"openai/gpt-oss-120b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://arxiv.org/abs/2508.10925","sourceLocator":"rank=51;row=50"}]},{"modelRef":"openai/gpt-oss-20b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://arxiv.org/abs/2508.10925","sourceLocator":"rank=53;row=52"}]},{"modelRef":"KAT-Coder-Pro-V1","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=102","observedAt":"2026-08-21"}]},{"modelRef":"amazon-nova-experimental-chat-10-20","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=144","observedAt":"2026-08-21"}]},{"modelRef":"amazon-nova-experimental-chat-11-10","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=114","observedAt":"2026-08-21"}]},{"modelRef":"amazon-nova-experimental-chat-12-10","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=508","observedAt":"2026-08-21"}]},{"modelRef":"amazon-nova-experimental-chat-26-02-10","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=43","observedAt":"2026-08-21"}]},{"modelRef":"athene-v2-chat","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=232","observedAt":"2026-08-21"}]},{"modelRef":"claude-opus-4-1-20250805-thinking-16k","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=106","observedAt":"2026-08-21"}]},{"modelRef":"command-a-03-2025","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=204","observedAt":"2026-08-21"}]},{"modelRef":"deepseek-r1","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=162","observedAt":"2026-08-21"}]},{"modelRef":"deepseek-v2.5","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=257","observedAt":"2026-08-21"}]},{"modelRef":"deepseek-v2.5-1210","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=231","observedAt":"2026-08-21"}]},{"modelRef":"deepseek-v3","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=202","observedAt":"2026-08-21"}]},{"modelRef":"deepseek-v3-0324","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=158","observedAt":"2026-08-21"}]},{"modelRef":"devstral-medium-2507","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=117","observedAt":"2026-08-21"}]},{"modelRef":"ernie-5.0-0110","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=439","observedAt":"2026-08-21"}]},{"modelRef":"ernie-5.0-preview-1022","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=433","observedAt":"2026-08-21"}]},{"modelRef":"ernie-5.0-preview-1203","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=455","observedAt":"2026-08-21"}]},{"modelRef":"glm-4-plus","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=235","observedAt":"2026-08-21"}]},{"modelRef":"glm-4.5","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=470","observedAt":"2026-08-21"}]},{"modelRef":"glm-4.5-air","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=150","observedAt":"2026-08-21"}]},{"modelRef":"gpt-5.3-chat-latest","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=148","observedAt":"2026-08-21"}]},{"modelRef":"gpt-oss-120b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=171","observedAt":"2026-08-21"}]},{"modelRef":"gpt-oss-20b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=237","observedAt":"2026-08-21"}]},{"modelRef":"grok-3-mini-beta","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=175","observedAt":"2026-08-21"}]},{"modelRef":"grok-3-mini-high","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=167","observedAt":"2026-08-21"}]},{"modelRef":"grok-3-preview-02-24","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=512","observedAt":"2026-08-21"}]},{"modelRef":"grok-4-fast-chat","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=125","observedAt":"2026-08-21"}]},{"modelRef":"grok-4.1","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=465","observedAt":"2026-08-21"}]},{"modelRef":"grok-code-fast-1","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-webdev"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=webdev;split=latest;row_idx=115","observedAt":"2026-08-21"}]},{"modelRef":"hunyuan-large-2025-02-10","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=236","observedAt":"2026-08-21"}]},{"modelRef":"hunyuan-t1-20250711","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=136","observedAt":"2026-08-21"}]},{"modelRef":"hunyuan-turbo-0110","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=219","observedAt":"2026-08-21"}]},{"modelRef":"hunyuan-turbos-20250226","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=212","observedAt":"2026-08-21"}]},{"modelRef":"hunyuan-turbos-20250416","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=155","observedAt":"2026-08-21"}]},{"modelRef":"intellect-3","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=181","observedAt":"2026-08-21"}]},{"modelRef":"internvl2-26b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=127","observedAt":"2026-08-21"}]},{"modelRef":"internvl2-4b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=139","observedAt":"2026-08-21"}]},{"modelRef":"kimi-k2-0711-preview","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=163","observedAt":"2026-08-21"}]},{"modelRef":"kimi-k2-0905-preview","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=153","observedAt":"2026-08-21"}]},{"modelRef":"ling-flash-2.0","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=174","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.1-405b-instruct-bf16","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=242","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.2-vision-11b-instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=136","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.2-vision-90b-instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=128","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.3-nemotron-49b-super-v1","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=221","observedAt":"2026-08-21"}]},{"modelRef":"llava-v1.6-34b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=141","observedAt":"2026-08-21"}]},{"modelRef":"longcat-flash-chat","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=524","observedAt":"2026-08-21"}]},{"modelRef":"longcat-flash-chat-2602-exp","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=477","observedAt":"2026-08-21"}]},{"modelRef":"magistral-medium-2506","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=271","observedAt":"2026-08-21"}]},{"modelRef":"minimax-m1","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=190","observedAt":"2026-08-21"}]},{"modelRef":"molmo-72b-0924","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=124","observedAt":"2026-08-21"}]},{"modelRef":"molmo-7b-d-0924","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=137","observedAt":"2026-08-21"}]},{"modelRef":"nova-2-lite","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=176","observedAt":"2026-08-21"}]},{"modelRef":"nvidia-nemotron-3-nano-30b-a3b-bf16","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=187","observedAt":"2026-08-21"}]},{"modelRef":"nvidia-nemotron-3-super-120b-a12b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=154","observedAt":"2026-08-21"}]},{"modelRef":"nvidia-nemotron-3-ultra-550b-a55b-nvfp4","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=442","observedAt":"2026-08-21"}]},{"modelRef":"o1-mini","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=216","observedAt":"2026-08-21"}]},{"modelRef":"o1-preview","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=184","observedAt":"2026-08-21"}]},{"modelRef":"o3-mini","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=215","observedAt":"2026-08-21"}]},{"modelRef":"o3-mini-high","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=198","observedAt":"2026-08-21"}]},{"modelRef":"olmo-3-32b-think","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=228","observedAt":"2026-08-21"}]},{"modelRef":"olmo-3.1-32b-instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=220","observedAt":"2026-08-21"}]},{"modelRef":"pixtral-12b-2409","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=125","observedAt":"2026-08-21"}]},{"modelRef":"pixtral-large-2411","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=113","observedAt":"2026-08-21"}]},{"modelRef":"qwen-plus-0125","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=209","observedAt":"2026-08-21"}]},{"modelRef":"qwen2-vl-72b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=120","observedAt":"2026-08-21"}]},{"modelRef":"qwen2-vl-7b-instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=132","observedAt":"2026-08-21"}]},{"modelRef":"qwen2.5-72b-instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=259","observedAt":"2026-08-21"}]},{"modelRef":"qwen2.5-max","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=168","observedAt":"2026-08-21"}]},{"modelRef":"qwen2.5-plus-1127","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=227","observedAt":"2026-08-21"}]},{"modelRef":"qwen2.5-vl-72b-instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=109","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-235b-a22b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=169","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-235b-a22b-instruct-2507","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=103","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-235b-a22b-no-thinking","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=145","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-235b-a22b-thinking-2507","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=117","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-30b-a3b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=217","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-30b-a3b-instruct-2507","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=149","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-32b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=193","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-max-2025-09-23","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=118","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-max-preview","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=446","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-next-80b-a3b-instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=105","observedAt":"2026-08-21"}]},{"modelRef":"qwen3-next-80b-a3b-thinking","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=166","observedAt":"2026-08-21"}]},{"modelRef":"qwq-32b","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=207","observedAt":"2026-08-21"}]},{"modelRef":"ring-flash-2.0","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=203","observedAt":"2026-08-21"}]},{"modelRef":"step-2-16k-exp-202412","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=210","observedAt":"2026-08-21"}]},{"modelRef":"step-3.5-flash","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=133","observedAt":"2026-08-21"}]},{"modelRef":"trinity-large-preview","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=194","observedAt":"2026-08-21"}]},{"modelRef":"yi-lightning","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=224","observedAt":"2026-08-21"}]},{"modelRef":"Baichuan-13B-Base","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bbh_external","epoch-gsm8k_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-07-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2309.10305","sourceLocator":"bbh_external.csv:row=37;column=Average","observedAt":"2023-07-11"}]},{"modelRef":"Composer 2.5","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-cursorbench_external","epoch-frontiercode_external","epoch-frontierswe_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-05-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"cursorbench_external.csv:row=54;column=Score","observedAt":"2026-05-18"}]},{"modelRef":"DeepSeek-Coder-V2-Base","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_ai2_external","epoch-gsm8k_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-06-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"arc_ai2_external.csv:row=106;column=Challenge score","observedAt":"2024-06-17"}]},{"modelRef":"GLM-4.5-Air","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-forecastbench_external","epoch-gso_external","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-07-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=45;column=Overall score","observedAt":"2025-07-20"}]},{"modelRef":"Hermes-2-Theta-Llama-3-70B","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-math_level_5","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-06-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=79;column=mean_score","observedAt":"2024-06-20"}]},{"modelRef":"Llama-3.1-Tulu-3-70B-DPO","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-math_level_5","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-11-21","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=104;column=mean_score","observedAt":"2024-11-21"}]},{"modelRef":"Meta-Llama-3-70B","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=607;column=ECI Score","observedAt":"2024-04-18"}]},{"modelRef":"Meta-Llama-3-8B","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=611;column=ECI Score","observedAt":"2024-04-18"}]},{"modelRef":"MiniMax-M2","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-terminalbench_external","epoch-vending_bench_2_external","epoch-webdev_arena_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-01","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"terminalbench_external.csv:row=136;column=Accuracy mean","observedAt":"2025-11-01"}]},{"modelRef":"Mistral-7B-Instruct-v0.3","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-math_level_5","epoch-mmlu_external","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-05-27","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=89;column=mean_score","observedAt":"2024-05-27"}]},{"modelRef":"Mixtral-8x22B-Instruct-v0.1","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=609;column=ECI Score","observedAt":"2024-04-17"}]},{"modelRef":"Phi-3.5-MoE-instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-08-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"bool_q_external.csv:row=121;column=Score","observedAt":"2024-08-17"}]},{"modelRef":"Phi-3.5-mini-instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-08-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"bool_q_external.csv:row=120;column=Score","observedAt":"2024-08-16"}]},{"modelRef":"Qwen/Qwen3-235B-A22B-Thinking-2507","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","frontiermath"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-07-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=215;column=ECI Score","observedAt":"2025-07-25"}]},{"modelRef":"Qwen3-32B","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["aider-polyglot","epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-08","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=25;column=Percent correct","observedAt":"2025-05-08"}]},{"modelRef":"Ring-2.6-1T","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-ale_bench_external","epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-05-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=94;column=Performance","observedAt":"2026-05-14"}]},{"modelRef":"T5-3B","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-common_sense_qa_2_external","epoch-superglue_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/pdf/1910.10683","sourceLocator":"bool_q_external.csv:row=206;column=Score"}]},{"modelRef":"T5-Base","rowCount":3,"numericRowCount":3,"benchmarkCount":2,"benchmarkIds":["epoch-bool_q_external","epoch-superglue_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/pdf/1910.10683","sourceLocator":"bool_q_external.csv:row=204;column=Score"}]},{"modelRef":"ada","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=87;column=Score"}]},{"modelRef":"babbage","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2020-06-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=88;column=Score","observedAt":"2020-06-22"}]},{"modelRef":"bloom","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-07-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=72;column=Score","observedAt":"2022-07-06"}]},{"modelRef":"c4ai-command-r-plus-08-2024","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-mmlu_external","epoch-simplebench_external","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-08-30","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/lite/latest/#/leaderboard/mmlu","sourceLocator":"mmlu_external.csv:row=36;column=EM","observedAt":"2024-08-30"}]},{"modelRef":"claude-3-7-sonnet-20250219_1K","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=199;column=Score","observedAt":"2025-02-24"}]},{"modelRef":"claude-3-7-sonnet-20250219_8K","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=185;column=Score","observedAt":"2025-02-24"}]},{"modelRef":"claude-sonnet-4-20250514_1K","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=187;column=Score","observedAt":"2025-05-22"}]},{"modelRef":"claude-sonnet-4-20250514_59K","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=238;column=ECI Score","observedAt":"2025-05-22"}]},{"modelRef":"claude-sonnet-4-20250514_8K","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=157;column=Score","observedAt":"2025-05-22"}]},{"modelRef":"computer-use-preview-2025-03-11","rowCount":3,"numericRowCount":3,"benchmarkCount":1,"benchmarkIds":["osworld"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://os-world.github.io/","sourceLocator":"os_world_external.csv:row=26;column=Score","observedAt":"2025-03-11"}]},{"modelRef":"curie","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=82;column=Score"}]},{"modelRef":"davinci","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=64;column=Score"}]},{"modelRef":"gemini-1.5-pro-001-feb24","rowCount":3,"numericRowCount":3,"benchmarkCount":2,"benchmarkIds":["epoch-bbh_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://storage.googleapis.com/deepmind-media/gemini/gemini_v1_5_report.pdf","sourceLocator":"bbh_external.csv:row=92;column=Average","observedAt":"2024-02-15"}]},{"modelRef":"gemini-2.5-flash-preview-04-17 (24K thinking)","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_external","epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arcprize.org/leaderboard","sourceLocator":"arc_agi_external.csv:row=172;column=Score","observedAt":"2025-04-17"}]},{"modelRef":"gemini-2.5-flash-preview-05-20_1K","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=155;column=Score","observedAt":"2025-05-20"}]},{"modelRef":"gemini-2.5-flash-preview-05-20_8K","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=156;column=Score","observedAt":"2025-05-20"}]},{"modelRef":"gemini-2.5-pro_8K","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=149;column=Score","observedAt":"2025-06-17"}]},{"modelRef":"gemini-3-flash-preview_low","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-deepresearchbench_external","epoch-epoch_capabilities_index","epoch-mystery_game_puzzles"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=12;column=Average score","observedAt":"2025-12-17"}]},{"modelRef":"gemini-3-flash-preview_minimal","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-deepresearchbench_external","epoch-epoch_capabilities_index","epoch-mystery_game_puzzles"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-12-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=16;column=Average score","observedAt":"2025-12-17"}]},{"modelRef":"gemini-3.1-pro-preview-customtools","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-vending_bench_2_external","swebench-verified"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-02-19","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=189;column=ECI Score","observedAt":"2026-02-19"}]},{"modelRef":"gemma-3-12b-it","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-critpt_external","epoch-mindcube_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=133;column=Accuracy","observedAt":"2025-03-12"}]},{"modelRef":"glm-4.5","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-ale_bench_external","epoch-lech_mazur_writing_external","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=99;column=Performance","observedAt":"2025-08-03"}]},{"modelRef":"gpt-4-32k-0314","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-hella_swag_external","epoch-wino_grande_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-03-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=422;column=ECI Score","observedAt":"2023-03-14"}]},{"modelRef":"gpt-4-turbo","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-metr_time_horizons_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=741;column=ECI Score","observedAt":"2023-11-06"}]},{"modelRef":"gpt-5.5-pro-pre-release_high","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","frontiermath"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=182;column=ECI Score","observedAt":"2026-04-23"}]},{"modelRef":"gpt-oss-120b_high","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=2;column=Percent correct","observedAt":"2025-08-05"}]},{"modelRef":"gpt-oss-120b_unknown","rowCount":3,"numericRowCount":3,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-terminalbench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=798;column=ECI Score","observedAt":"2025-08-05"}]},{"modelRef":"gpt-oss-20b_unknown","rowCount":3,"numericRowCount":3,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-terminalbench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=799;column=ECI Score","observedAt":"2025-08-05"}]},{"modelRef":"grok-3-mini_low","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=197;column=Score","observedAt":"2025-06-24"}]},{"modelRef":"grok-4-0709_high","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index","epoch-gdpval_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-07-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=74;column=Percent correct","observedAt":"2025-07-11"}]},{"modelRef":"grok-build-0.1","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-critpt_external","epoch-gbaeval_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-05-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=48;column=Accuracy","observedAt":"2026-05-29"}]},{"modelRef":"inkling-small_high","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=88;column=Score","observedAt":"2026-07-15"}]},{"modelRef":"inkling-small_low","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=131;column=Score","observedAt":"2026-07-15"}]},{"modelRef":"inkling-small_medium","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=104;column=Score","observedAt":"2026-07-15"}]},{"modelRef":"inkling-small_minimal","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=143;column=Score","observedAt":"2026-07-15"}]},{"modelRef":"inkling-small_none","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=217;column=Score","observedAt":"2026-07-15"}]},{"modelRef":"laguna-m.1","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-proofbench_external","epoch-surface_evolver_bench_external","epoch-webdev_arena_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=59;column=Accuracy","observedAt":"2026-04-28"}]},{"modelRef":"mpt-30b-instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-epoch_capabilities_index","epoch-gsm8k_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-06-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=35;column=Score","observedAt":"2023-06-22"}]},{"modelRef":"o3-pro-2025-06-10_low","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-10","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=160;column=Score","observedAt":"2025-06-10"}]},{"modelRef":"open-mistral-nemo-2407","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-07-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=325;column=ECI Score","observedAt":"2024-07-18"}]},{"modelRef":"open-mixtral-8x22b","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=321;column=ECI Score","observedAt":"2024-04-17"}]},{"modelRef":"open-mixtral-8x7b","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-12-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=323;column=ECI Score","observedAt":"2023-12-11"}]},{"modelRef":"openai/gpt-oss-120b_low","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-critpt_external","epoch-epoch_capabilities_index","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=140;column=Accuracy","observedAt":"2025-08-05"}]},{"modelRef":"opt-175b","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-05-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=46;column=Score","observedAt":"2022-05-02"}]},{"modelRef":"opt-66b","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-05-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=55;column=Score","observedAt":"2022-05-03"}]},{"modelRef":"qwen-plus-2025-01-25","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-math_level_5","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=30;column=mean_score","observedAt":"2025-01-25"}]},{"modelRef":"qwen-turbo-2024-11-01","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-math_level_5","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-11-01","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=29;column=mean_score","observedAt":"2024-11-01"}]},{"modelRef":"qwen2.5-32b-instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-math_level_5","epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=56;column=mean_score","observedAt":"2024-09-17"}]},{"modelRef":"qwen2.5-max","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-lech_mazur_writing_external","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=520;column=ECI Score","observedAt":"2025-01-28"}]},{"modelRef":"step-3.7-flash","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-ale_bench_external","epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-05-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=72;column=Performance","observedAt":"2026-05-29"}]},{"modelRef":"text-ada-001","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=89;column=Score"}]},{"modelRef":"text-babbage-001","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=91;column=Score"}]},{"modelRef":"text-curie-001","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external","epoch-hella_swag_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=86;column=Score"}]},{"modelRef":"zai-org/GLM-4.6","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-frontiermath_tier_4","frontiermath"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-30","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=222;column=ECI Score","observedAt":"2025-09-30"}]},{"modelRef":"Claude 2","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["swebench-lite","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2023-10-10","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[79]","observedAt":"2023-10-10"}]},{"modelRef":"Claude 3.5-Sonnet-20241022","rowCount":3,"numericRowCount":3,"benchmarkCount":2,"benchmarkIds":["swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-01-22","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[2]=Test;results[7]","observedAt":"2024-11-21"}]},{"modelRef":"Claude Sonnet 4","rowCount":3,"numericRowCount":3,"benchmarkCount":2,"benchmarkIds":["swebench-multimodal","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-09-24","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[5]=Multimodal;results[3]","observedAt":"2025-05-28"}]},{"modelRef":"CodeAct v2.1 (claude-3-5-sonnet-20241022)","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["swebench-lite","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-11-03","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[22]","observedAt":"2024-10-25"}]},{"modelRef":"GPT-3.5","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["swebench-lite","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2023-10-10","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[83]","observedAt":"2023-10-10"}]},{"modelRef":"Gemini 3 Flash","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["swebench-bash-only","swebench-multilingual","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2026-02-17","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[1]","observedAt":"2026-02-17"}]},{"modelRef":"Gemini 3 Pro","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["swebench-bash-only","swebench-multilingual","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2026-02-26","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[14]","observedAt":"2026-02-26"}]},{"modelRef":"Gemini 3 Pro Preview","rowCount":3,"numericRowCount":3,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-11-20","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[5]","observedAt":"2025-11-18"}]},{"modelRef":"Kimi K2 Instruct","rowCount":3,"numericRowCount":3,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-08-07","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[36]","observedAt":"2025-08-07"}]},{"modelRef":"SWE-Llama 13B","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["swebench-lite","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2023-10-10","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[82]","observedAt":"2023-10-10"}]},{"modelRef":"SWE-Llama 7B","rowCount":3,"numericRowCount":3,"benchmarkCount":3,"benchmarkIds":["swebench-lite","swebench-test","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":3},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2023-10-10","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[81]","observedAt":"2023-10-10"}]},{"modelRef":"openpangu/openPangu-2.0-Flash","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://huggingface.co/openpangu/openPangu-2.0-Flash","sourceLocator":"rank=15;row=14"}]},{"modelRef":"RedHatAI/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=55;row=54"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-MXFP4_MOE-dequant-bf16-vllm","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=58;row=57"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q3_K_M-dequant-bf16-vllm","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=56;row=55"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_M-dequant-bf16-vllm","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=57;row=56"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=54;row=53"}]},{"modelRef":"openpangu/openPangu-2.0-Flash","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/openpangu/openPangu-2.0-Flash","sourceLocator":"rank=49;row=48"}]},{"modelRef":"INCModel/Kimi-K2.6-MXFP4-CT-AutoRound","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6","sourceLocator":"rank=30;row=29"}]},{"modelRef":"RedHatAI/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=53;row=52"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-MXFP4_MOE-dequant-bf16-vllm","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=56;row=55"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q3_K_M-dequant-bf16-vllm","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=54;row=53"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_M-dequant-bf16-vllm","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=55;row=54"}]},{"modelRef":"moonshotai/Kimi-K2-Thinking","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2-Thinking","sourceLocator":"rank=19;row=18"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=52;row=51"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16","sourceLocator":"rank=25;row=24"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4","sourceLocator":"rank=26;row=25"}]},{"modelRef":"ornith-ai/Ornith-1.5-35B-A3B","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B","sourceLocator":"rank=32;row=31"}]},{"modelRef":"ornith-ai/Ornith-1.5-397B","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-397B","sourceLocator":"rank=1;row=0"}]},{"modelRef":"ornith-ai/Ornith-1.5-9B","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-9B","sourceLocator":"rank=37;row=36"}]},{"modelRef":"openpangu/openPangu-2.0-Flash","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/openpangu/openPangu-2.0-Flash","sourceLocator":"rank=50;row=49"}]},{"modelRef":"RWKV-4-Raven-14B","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=385","observedAt":"2026-08-21"}]},{"modelRef":"alpaca-13b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=387","observedAt":"2026-08-21"}]},{"modelRef":"amazon-nova-experimental-chat-10-09","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=172","observedAt":"2026-08-21"}]},{"modelRef":"amazon-nova-experimental-chat-26-01-10","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=137","observedAt":"2026-08-21"}]},{"modelRef":"amazon-nova-micro-v1.0","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=292","observedAt":"2026-08-21"}]},{"modelRef":"athene-70b-0725","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=262","observedAt":"2026-08-21"}]},{"modelRef":"c4ai-aya-expanse-32b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=286","observedAt":"2026-08-21"}]},{"modelRef":"c4ai-aya-expanse-8b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=306","observedAt":"2026-08-21"}]},{"modelRef":"c4ai-aya-vision-32b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=130","observedAt":"2026-08-21"}]},{"modelRef":"chatglm-6b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=388","observedAt":"2026-08-21"}]},{"modelRef":"chatglm3-6b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=382","observedAt":"2026-08-21"}]},{"modelRef":"codellama-34b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=361","observedAt":"2026-08-21"}]},{"modelRef":"cogvlm2-llama3-chat-19b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=142","observedAt":"2026-08-21"}]},{"modelRef":"command-r","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=315","observedAt":"2026-08-21"}]},{"modelRef":"command-r-08-2024","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=302","observedAt":"2026-08-21"}]},{"modelRef":"command-r-plus","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=295","observedAt":"2026-08-21"}]},{"modelRef":"command-r-plus-08-2024","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=279","observedAt":"2026-08-21"}]},{"modelRef":"dbrx-instruct-preview","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=334","observedAt":"2026-08-21"}]},{"modelRef":"deepseek-coder-v2","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=300","observedAt":"2026-08-21"}]},{"modelRef":"deepseek-llm-67b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=342","observedAt":"2026-08-21"}]},{"modelRef":"dolly-v2-12b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=392","observedAt":"2026-08-21"}]},{"modelRef":"fastchat-t5-3b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=390","observedAt":"2026-08-21"}]},{"modelRef":"gemini-advanced-0514","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=251","observedAt":"2026-08-21"}]},{"modelRef":"gemini-pro","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=327","observedAt":"2026-08-21"}]},{"modelRef":"gemini-pro-dev-api","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=321","observedAt":"2026-08-21"}]},{"modelRef":"gemma-1.1-2b-it","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=378","observedAt":"2026-08-21"}]},{"modelRef":"gemma-1.1-7b-it","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=348","observedAt":"2026-08-21"}]},{"modelRef":"gemma-2-27b-it","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=277","observedAt":"2026-08-21"}]},{"modelRef":"gemma-2-2b-it","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=319","observedAt":"2026-08-21"}]},{"modelRef":"gemma-2-9b-it","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=293","observedAt":"2026-08-21"}]},{"modelRef":"gemma-2-9b-it-simpo","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=282","observedAt":"2026-08-21"}]},{"modelRef":"gemma-2b-it","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=379","observedAt":"2026-08-21"}]},{"modelRef":"gemma-3n-e4b-it","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=222","observedAt":"2026-08-21"}]},{"modelRef":"gemma-7b-it","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=365","observedAt":"2026-08-21"}]},{"modelRef":"glm-4-0520","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=283","observedAt":"2026-08-21"}]},{"modelRef":"glm-4-plus-0111","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=205","observedAt":"2026-08-21"}]},{"modelRef":"gpt-3.5-turbo-0125","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=331","observedAt":"2026-08-21"}]},{"modelRef":"gpt-3.5-turbo-1106","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=347","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4-0125-preview","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=265","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4-0314","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=294","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4-0613","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=305","observedAt":"2026-08-21"}]},{"modelRef":"gpt-4-1106-preview","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=263","observedAt":"2026-08-21"}]},{"modelRef":"granite-3.0-2b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=356","observedAt":"2026-08-21"}]},{"modelRef":"granite-3.0-8b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=346","observedAt":"2026-08-21"}]},{"modelRef":"granite-3.1-2b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=330","observedAt":"2026-08-21"}]},{"modelRef":"granite-3.1-8b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=320","observedAt":"2026-08-21"}]},{"modelRef":"grok-2-2024-08-13","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=223","observedAt":"2026-08-21"}]},{"modelRef":"grok-2-mini-2024-08-13","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=249","observedAt":"2026-08-21"}]},{"modelRef":"hunyuan-standard-2025-02-10","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=254","observedAt":"2026-08-21"}]},{"modelRef":"hunyuan-standard-256k","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=297","observedAt":"2026-08-21"}]},{"modelRef":"hunyuan-standard-vision-2024-12-31","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=129","observedAt":"2026-08-21"}]},{"modelRef":"ibm-granite-h-small","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=273","observedAt":"2026-08-21"}]},{"modelRef":"internlm2_5-20b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=318","observedAt":"2026-08-21"}]},{"modelRef":"jamba-1.5-large","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=275","observedAt":"2026-08-21"}]},{"modelRef":"jamba-1.5-mini","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=303","observedAt":"2026-08-21"}]},{"modelRef":"koala-13b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=381","observedAt":"2026-08-21"}]},{"modelRef":"llama-2-13b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=351","observedAt":"2026-08-21"}]},{"modelRef":"llama-2-70b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=335","observedAt":"2026-08-21"}]},{"modelRef":"llama-2-7b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=368","observedAt":"2026-08-21"}]},{"modelRef":"llama-3-70b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=287","observedAt":"2026-08-21"}]},{"modelRef":"llama-3-8b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=312","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.1-405b-instruct-fp8","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=247","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.1-70b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=267","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.1-8b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=304","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.1-nemotron-51b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=281","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.1-nemotron-70b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=243","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.1-nemotron-ultra-253b-v1","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=213","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.1-tulu-3-70b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=269","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.1-tulu-3-8b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=299","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.2-1b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=366","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.2-3b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=338","observedAt":"2026-08-21"}]},{"modelRef":"llama-3.3-70b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=253","observedAt":"2026-08-21"}]},{"modelRef":"llava-onevision-qwen2-72b-ov","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=143","observedAt":"2026-08-21"}]},{"modelRef":"mercury","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=246","observedAt":"2026-08-21"}]},{"modelRef":"minicpm-v-2_6","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=144","observedAt":"2026-08-21"}]},{"modelRef":"ministral-8b-2410","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=301","observedAt":"2026-08-21"}]},{"modelRef":"mistral-7b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=377","observedAt":"2026-08-21"}]},{"modelRef":"mistral-7b-instruct-v0.2","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=350","observedAt":"2026-08-21"}]},{"modelRef":"mistral-large-2402","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=307","observedAt":"2026-08-21"}]},{"modelRef":"mistral-large-2407","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=260","observedAt":"2026-08-21"}]},{"modelRef":"mistral-large-2411","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=261","observedAt":"2026-08-21"}]},{"modelRef":"mistral-medium","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=313","observedAt":"2026-08-21"}]},{"modelRef":"mistral-small-24b-instruct-2501","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=276","observedAt":"2026-08-21"}]},{"modelRef":"mixtral-8x22b-instruct-v0.1","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=316","observedAt":"2026-08-21"}]},{"modelRef":"mixtral-8x7b-instruct-v0.1","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=326","observedAt":"2026-08-21"}]},{"modelRef":"mpt-7b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=384","observedAt":"2026-08-21"}]},{"modelRef":"nemotron-4-340b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=285","observedAt":"2026-08-21"}]},{"modelRef":"nvidia-llama-3.3-nemotron-super-49b-v1.5","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=197","observedAt":"2026-08-21"}]},{"modelRef":"nvidia-nemotron-3.5-lightning-30b-a3b-nvfp4","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=208","observedAt":"2026-08-21"}]},{"modelRef":"nvila-internal-15b-v1","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=140","observedAt":"2026-08-21"}]},{"modelRef":"oasst-pythia-12b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=389","observedAt":"2026-08-21"}]},{"modelRef":"olmo-2-0325-32b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=290","observedAt":"2026-08-21"}]},{"modelRef":"olmo-3.1-32b-think","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=258","observedAt":"2026-08-21"}]},{"modelRef":"olmo-7b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=374","observedAt":"2026-08-21"}]},{"modelRef":"openchat-3.5","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=345","observedAt":"2026-08-21"}]},{"modelRef":"openchat-3.5-0106","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=340","observedAt":"2026-08-21"}]},{"modelRef":"openhermes-2.5-mistral-7b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=349","observedAt":"2026-08-21"}]},{"modelRef":"palm-2","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=376","observedAt":"2026-08-21"}]},{"modelRef":"phi-3-medium-4k-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=323","observedAt":"2026-08-21"}]},{"modelRef":"phi-3-mini-128k-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=371","observedAt":"2026-08-21"}]},{"modelRef":"phi-3-mini-4k-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=358","observedAt":"2026-08-21"}]},{"modelRef":"phi-3-mini-4k-instruct-june-2024","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=355","observedAt":"2026-08-21"}]},{"modelRef":"phi-3-small-8k-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=337","observedAt":"2026-08-21"}]},{"modelRef":"phi-3-vision-128k-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=146","observedAt":"2026-08-21"}]},{"modelRef":"phi-3.5-vision-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=145","observedAt":"2026-08-21"}]},{"modelRef":"phi-4","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=291","observedAt":"2026-08-21"}]},{"modelRef":"qwen-14b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=370","observedAt":"2026-08-21"}]},{"modelRef":"qwen-max-0919","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=245","observedAt":"2026-08-21"}]},{"modelRef":"qwen-vl-max-1119","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=119","observedAt":"2026-08-21"}]},{"modelRef":"qwen1.5-110b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=308","observedAt":"2026-08-21"}]},{"modelRef":"qwen1.5-14b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=329","observedAt":"2026-08-21"}]},{"modelRef":"qwen1.5-32b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=324","observedAt":"2026-08-21"}]},{"modelRef":"qwen1.5-4b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=380","observedAt":"2026-08-21"}]},{"modelRef":"qwen1.5-72b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=311","observedAt":"2026-08-21"}]},{"modelRef":"qwen1.5-7b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=352","observedAt":"2026-08-21"}]},{"modelRef":"qwen2-72b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=296","observedAt":"2026-08-21"}]},{"modelRef":"qwen2.5-coder-32b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=278","observedAt":"2026-08-21"}]},{"modelRef":"qwen2.5-vl-32b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=611","observedAt":"2026-08-21"}]},{"modelRef":"qwq-32b-preview","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=317","observedAt":"2026-08-21"}]},{"modelRef":"reka-core-20240904","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=272","observedAt":"2026-08-21"}]},{"modelRef":"reka-flash-20240904","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=289","observedAt":"2026-08-21"}]},{"modelRef":"reka-flash-21b-20240226","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=314","observedAt":"2026-08-21"}]},{"modelRef":"reka-flash-21b-20240226-online","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=310","observedAt":"2026-08-21"}]},{"modelRef":"snowflake-arctic-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=343","observedAt":"2026-08-21"}]},{"modelRef":"starling-lm-7b-alpha","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=339","observedAt":"2026-08-21"}]},{"modelRef":"starling-lm-7b-beta","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=325","observedAt":"2026-08-21"}]},{"modelRef":"step-1o-vision-32k-highres","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=106","observedAt":"2026-08-21"}]},{"modelRef":"step-1v-32k","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=121","observedAt":"2026-08-21"}]},{"modelRef":"tulu-2-dpo-70b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=332","observedAt":"2026-08-21"}]},{"modelRef":"vicuna-13b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=363","observedAt":"2026-08-21"}]},{"modelRef":"vicuna-33b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=341","observedAt":"2026-08-21"}]},{"modelRef":"vicuna-7b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=375","observedAt":"2026-08-21"}]},{"modelRef":"wizardlm-13b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=357","observedAt":"2026-08-21"}]},{"modelRef":"wizardlm-70b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=333","observedAt":"2026-08-21"}]},{"modelRef":"yi-1.5-34b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=309","observedAt":"2026-08-21"}]},{"modelRef":"yi-34b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=328","observedAt":"2026-08-21"}]},{"modelRef":"yi-vision","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-vision"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=vision;split=latest;row_idx=135","observedAt":"2026-08-21"}]},{"modelRef":"zephyr-7b-beta","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=359","observedAt":"2026-08-21"}]},{"modelRef":"zephyr-orpo-141b-A35b-v0.1","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=322","observedAt":"2026-08-21"}]},{"modelRef":"Qwen2.5-Coder-32B-Instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-26","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=209;dirname=2024-12-22-13-22-32--polyglot-qwen-diff","observedAt":"2024-12-22"}]},{"modelRef":"GPT-4.1-2025-04-14","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=19;rank=20;model=GPT-4.1-2025-04-14 (FC);mode=FC"}]},{"modelRef":"GPT-4.1-mini-2025-04-14","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=26;rank=27;model=GPT-4.1-mini-2025-04-14 (FC);mode=FC"}]},{"modelRef":"GPT-4.1-nano-2025-04-14","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=57;rank=58;model=GPT-4.1-nano-2025-04-14 (FC);mode=FC"}]},{"modelRef":"Gemini-2.5-Flash-Lite","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=51;rank=52;model=Gemini-2.5-Flash-Lite (FC);mode=FC"}]},{"modelRef":"Gemini-3-Pro-Preview","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=2;rank=3;model=Gemini-3-Pro-Preview (Prompt);mode=Prompt"}]},{"modelRef":"Grok-4-0709","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=8;rank=9;model=Grok-4-0709 (Prompt);mode=Prompt"}]},{"modelRef":"Mistral-Medium-2505","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=47;rank=48;model=Mistral-Medium-2505;mode=unknown"}]},{"modelRef":"Open-Mistral-Nemo-2407","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=101;rank=102;model=Open-Mistral-Nemo-2407 (Prompt);mode=Prompt"}]},{"modelRef":"Qwen3-0.6B","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=91;rank=92;model=Qwen3-0.6B (FC);mode=FC"}]},{"modelRef":"Qwen3-14B","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=42;rank=43;model=Qwen3-14B (FC);mode=FC"}]},{"modelRef":"Qwen3-235B-A22B-Instruct-2507","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=22;rank=23;model=Qwen3-235B-A22B-Instruct-2507 (Prompt);mode=Prompt"}]},{"modelRef":"Qwen3-30B-A3B-Instruct-2507","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=40;rank=41;model=Qwen3-30B-A3B-Instruct-2507 (FC);mode=FC"}]},{"modelRef":"Qwen3-32B","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=28;rank=29;model=Qwen3-32B (FC);mode=FC"}]},{"modelRef":"Qwen3-4B-Instruct-2507","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=53;rank=54;model=Qwen3-4B-Instruct-2507 (FC);mode=FC"}]},{"modelRef":"Qwen3-8B","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=38;rank=39;model=Qwen3-8B (FC);mode=FC"}]},{"modelRef":"mistral-large-2411","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=45;rank=46;model=mistral-large-2411 (FC);mode=FC"}]},{"modelRef":"o3-2025-04-16","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=29;rank=30;model=o3-2025-04-16 (FC);mode=FC"}]},{"modelRef":"o4-mini-2025-04-16","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=20;rank=21;model=o4-mini-2025-04-16 (FC);mode=FC"}]},{"modelRef":"DeepSeek-R1-Distill-Qwen-14B","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=38;column=mean_score","observedAt":"2025-01-20"}]},{"modelRef":"DeepSeek-R1-Distill-Qwen-32B","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-balrog_external","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=24;column=Average progress","observedAt":"2025-01-20"}]},{"modelRef":"DeepSeek-V3-Base","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-26","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=856;column=ECI Score","observedAt":"2024-12-26"}]},{"modelRef":"DeepSeek-V3.2-Speciale","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-simplebench_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-12-01","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=50;column=Score (AVG@5)","observedAt":"2025-12-01"}]},{"modelRef":"Inkling_high","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=823;column=ECI Score","observedAt":"2026-07-15"}]},{"modelRef":"Kimi-K2-Instruct-0905","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-forecastbench_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=36;column=Overall score","observedAt":"2025-09-05"}]},{"modelRef":"Llama-3.2-11B-Vision-Instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-balrog_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=28;column=Average progress","observedAt":"2024-09-24"}]},{"modelRef":"Mistral-7B-Instruct-v0.1","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-piqa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-27","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=857;column=ECI Score","observedAt":"2023-09-27"}]},{"modelRef":"Mistral-Nemo-Instruct-2407","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-balrog_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-07-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=26;column=Average progress","observedAt":"2024-07-18"}]},{"modelRef":"Mixtral-8x22B-v0.1","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=827;column=ECI Score","observedAt":"2024-04-17"}]},{"modelRef":"Phi-4-mini-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-27","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=129;column=Accuracy","observedAt":"2025-02-27"}]},{"modelRef":"QwQ-32B","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["aider-polyglot","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=51;column=Percent correct","observedAt":"2025-03-06"}]},{"modelRef":"QwQ-32B-Preview","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-forecastbench_external","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-11-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=57;column=Overall score","observedAt":"2024-11-28"}]},{"modelRef":"Qwen-VL-Chat","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-08-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"science_qa_external.csv:row=88;column=Score","observedAt":"2023-08-20"}]},{"modelRef":"Qwen2.5-Coder-7B-Instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-gsm8k_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=752;column=ECI Score","observedAt":"2024-09-17"}]},{"modelRef":"Qwen3-30B-A3B","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-lech_mazur_writing_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://github.com/lechmazur/Writing","sourceLocator":"lech_mazur_writing_external.csv:row=16;column=Mean score","observedAt":"2025-04-29"}]},{"modelRef":"T5-Small","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-bool_q_external","epoch-superglue_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arxiv.org/pdf/1910.10683","sourceLocator":"bool_q_external.csv:row=203;column=Score"}]},{"modelRef":"WizardLM-2-8x22B","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=92;column=mean_score","observedAt":"2024-04-15"}]},{"modelRef":"Yi-1.5-34B-Chat","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-05-13","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=83;column=mean_score","observedAt":"2024-05-13"}]},{"modelRef":"amazon.nova-micro-v1:0","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-mmlu_external","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/lite/latest/#/leaderboard/mmlu","sourceLocator":"mmlu_external.csv:row=3;column=EM","observedAt":"2024-12-03"}]},{"modelRef":"c4ai-command-r-08-2024","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-mmlu_external","livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-08-30","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/lite/latest/#/leaderboard/mmlu","sourceLocator":"mmlu_external.csv:row=35;column=EM","observedAt":"2024-08-30"}]},{"modelRef":"chutes/Llama-4-Maverick-17B-128E-Instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=104;column=Accuracy","observedAt":"2025-04-06"}]},{"modelRef":"claude-1.3","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-mmlu_external","epoch-trivia_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-04-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://www-cdn.anthropic.com/5c49cc247484cecf107c699baf29250302e5da70/claude-2-model-card.pdf","sourceLocator":"mmlu_external.csv:row=21;column=EM","observedAt":"2023-04-18"}]},{"modelRef":"claude-3-7-sonnet-20250219_12K","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-simplebench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=794;column=ECI Score","observedAt":"2025-02-24"}]},{"modelRef":"claude-3-7-sonnet-20250219_15K","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-geobench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=801;column=ECI Score","observedAt":"2025-02-24"}]},{"modelRef":"claude-3-7-sonnet-20250219_2K","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-deepresearchbench_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://drb.futuresearch.ai","sourceLocator":"deepresearchbench_external.csv:row=33;column=Average score","observedAt":"2025-02-24"}]},{"modelRef":"claude-opus-4-1-20250805_24K","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-mystery_game_puzzles"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=130;column=ECI Score","observedAt":"2025-08-05"}]},{"modelRef":"claude-sonnet-4-20250514_12K","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-simplebench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=796;column=ECI Score","observedAt":"2025-05-22"}]},{"modelRef":"claude-sonnet-4-20250514_2K","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-deepresearchbench_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://drb.futuresearch.ai","sourceLocator":"deepresearchbench_external.csv:row=28;column=Average score","observedAt":"2025-05-22"}]},{"modelRef":"codex-mini-2025-05-16","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-05-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=177;column=Score","observedAt":"2025-05-16"}]},{"modelRef":"cogito-671b-v2.1","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-19","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=139;column=Accuracy","observedAt":"2025-11-19"}]},{"modelRef":"command-a-plus-05-2026","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-05-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=101;column=Accuracy","observedAt":"2026-05-20"}]},{"modelRef":"dbrx-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-03-27","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=93;column=mean_score","observedAt":"2024-03-27"}]},{"modelRef":"deepseek-llm-67b-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-11-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=91;column=mean_score","observedAt":"2023-11-29"}]},{"modelRef":"devstral-small-2512","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-12-09","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=123;column=Accuracy","observedAt":"2025-12-09"}]},{"modelRef":"falcon-40b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-gsm8k_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-05-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=745;column=ECI Score","observedAt":"2023-05-25"}]},{"modelRef":"gemini-1.5-flash-0514","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-05-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=825;column=ECI Score","observedAt":"2024-05-14"}]},{"modelRef":"gemini-1.5-flash-8b-001","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-otis_mock_aime_2024_2025","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-10-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"otis_mock_aime_2024_2025.csv:row=219;column=mean_score","observedAt":"2024-10-03"}]},{"modelRef":"gemini-2.0-flash-02-05","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-enigma_eval_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=41;column=Accuracy","observedAt":"2025-02-05"}]},{"modelRef":"gemini-2.5-deep-think-2025-08-01-webapp","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-frontiermath_tier_4","frontiermath"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-01","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=73;column=mean_score","observedAt":"2025-08-01"}]},{"modelRef":"gemini-2.5-flash-lite-preview-06-17-thinking","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-ale_bench_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=102;column=Performance","observedAt":"2025-06-17"}]},{"modelRef":"gemini-2.5-flash-preview-04-17 (16K thinking)","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=817;column=ECI Score","observedAt":"2025-04-17"}]},{"modelRef":"gemini-2.5-flash-preview-09-2025_16k","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=815;column=ECI Score","observedAt":"2025-09-25"}]},{"modelRef":"gemini-2.5-pro-preview-06-05_32K","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["aider-polyglot","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=18;column=Percent correct","observedAt":"2025-06-06"}]},{"modelRef":"gemini-3-pro-preview_low","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-deepresearchbench_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"deepresearchbench_external.csv:row=29;column=Average score","observedAt":"2025-11-18"}]},{"modelRef":"glm-4.5_thinking","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-algotune_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=10;column=Score","observedAt":"2025-08-05"}]},{"modelRef":"gpt-5-chat","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=818;column=ECI Score"}]},{"modelRef":"gpt-5-codex_high","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-apex_agents_external","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=35;column=Pass@1 score","observedAt":"2025-09-15"}]},{"modelRef":"gpt-5-pro-2025-10-06_medium","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-algotune_external","epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-10-07","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"algotune_external.csv:row=19;column=Score","observedAt":"2025-10-07"}]},{"modelRef":"gpt-oss-120b_medium","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=816;column=ECI Score","observedAt":"2025-08-05"}]},{"modelRef":"gpt-oss-20b_medium","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=821;column=ECI Score","observedAt":"2025-08-05"}]},{"modelRef":"granite-4.1-30b","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=131;column=Accuracy","observedAt":"2026-04-29"}]},{"modelRef":"grok-3-mini_high","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=814;column=ECI Score","observedAt":"2025-06-24"}]},{"modelRef":"grok-4-1","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-apex_agents_external","epoch-webdev_arena_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"apex_agents_external.csv:row=48;column=Pass@1 score","observedAt":"2025-11-17"}]},{"modelRef":"kimi-k2-0711-preview","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-07-12","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=819;column=ECI Score","observedAt":"2025-07-12"}]},{"modelRef":"laguna-xs.2","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-proofbench_external","epoch-webdev_arena_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"proofbench_external.csv:row=58;column=Accuracy","observedAt":"2026-04-28"}]},{"modelRef":"llava-v1.6-vicuna-7b","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-01-31","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"science_qa_external.csv:row=86;column=Score","observedAt":"2024-01-31"}]},{"modelRef":"magistral-medium-2509","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=99;column=Accuracy","observedAt":"2025-09-18"}]},{"modelRef":"magistral-small-2509","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=100;column=Accuracy","observedAt":"2025-09-18"}]},{"modelRef":"ministral-3b-2410","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-10-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=95;column=mean_score","observedAt":"2024-10-16"}]},{"modelRef":"ministral-8b-2410","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-10-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=97;column=mean_score","observedAt":"2024-10-16"}]},{"modelRef":"mistral-small-2506","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=132;column=Accuracy","observedAt":"2025-06-20"}]},{"modelRef":"moonshotai/kimi-k2-0905","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["aider-polyglot","epoch-ale_bench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"aider_polyglot_external.csv:row=75;column=Percent correct","observedAt":"2025-07-17"}]},{"modelRef":"muse-spark-1.1_high","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-surface_evolver_bench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-09","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=400;column=ECI Score","observedAt":"2026-07-09"}]},{"modelRef":"nova-2.0-pro-preview_low","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-12-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=137;column=Accuracy","observedAt":"2025-12-02"}]},{"modelRef":"nova-2.0-pro-preview_medium","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-12-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=125;column=Accuracy","observedAt":"2025-12-02"}]},{"modelRef":"nova-2.0-pro-preview_none","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-12-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=136;column=Accuracy","observedAt":"2025-12-02"}]},{"modelRef":"o1-pro-2025-03-19","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-enigma_eval_external","hle"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-19","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=19;column=Accuracy","observedAt":"2025-03-19"}]},{"modelRef":"o3-mini-2025-01-31_unknown","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-31","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=600;column=ECI Score","observedAt":"2025-01-31"}]},{"modelRef":"o4-mini-2025-04-16_unknown","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-epoch_capabilities_index","epoch-forecastbench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=594;column=ECI Score","observedAt":"2025-04-16"}]},{"modelRef":"open-mistral-7b","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-math_level_5","gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-27","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"math_level_5.csv:row=96;column=mean_score","observedAt":"2023-09-27"}]},{"modelRef":"qwen2.5-7b-instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-balrog_external","epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-26","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=35;column=Average progress","observedAt":"2025-02-26"}]},{"modelRef":"solar-pro-3","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-critpt_external","epoch-scicode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-01-27","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"critpt_external.csv:row=128;column=Accuracy","observedAt":"2026-01-27"}]},{"modelRef":"tiny-recursion-model","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-arc_agi_2_external","epoch-arc_agi_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-10-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"arc_agi_2_external.csv:row=122;column=Score","observedAt":"2025-10-06"}]},{"modelRef":"vicuna-13b-v1.3","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["epoch-bool_q_external","epoch-gsm8k_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-06-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/boolq","sourceLocator":"bool_q_external.csv:row=44;column=Score","observedAt":"2023-06-18"}]},{"modelRef":"Claude Sonnet 3.5","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["swebench-multimodal"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-10-06","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[5]=Multimodal;results[13]","observedAt":"2024-10-06"}]},{"modelRef":"Devstral (2512)","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-12-09","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[32]","observedAt":"2025-12-09"}]},{"modelRef":"Devstral Small (2512)","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-12-09","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[27]","observedAt":"2025-12-09"}]},{"modelRef":"GLM 4.5","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-08-22","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[31]","observedAt":"2025-08-22"}]},{"modelRef":"GLM 4.6","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-12-01","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[30]","observedAt":"2025-12-01"}]},{"modelRef":"GPT 4.1","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-07-26","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[37]","observedAt":"2025-07-26"}]},{"modelRef":"GPT 4.1 mini","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-07-20","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[41]","observedAt":"2025-07-20"}]},{"modelRef":"GPT 4o","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-07-20","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[42]","observedAt":"2025-07-20"}]},{"modelRef":"Gemini 2.0 Flash","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-07-26","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[44]","observedAt":"2025-07-26"}]},{"modelRef":"Kimi K2","rowCount":2,"numericRowCount":2,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-10-14","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[34]","observedAt":"2025-10-14"}]},{"modelRef":"Kimi K2 Thinking","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-12-10","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[22]","observedAt":"2025-12-10"}]},{"modelRef":"Llama 4 Maverick Instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-07-20","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[43]","observedAt":"2025-07-20"}]},{"modelRef":"Llama 4 Scout Instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-07-20","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[45]","observedAt":"2025-07-20"}]},{"modelRef":"MCTS Refine 7B","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-lite","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-06-27","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[76]","observedAt":"2025-06-27"}]},{"modelRef":"MiniMax M2","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-11-24","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[23]","observedAt":"2025-11-24"}]},{"modelRef":"O3 Mini","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-lite","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-02-14","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[43]","observedAt":"2025-02-14"}]},{"modelRef":"Qwen2.5-Coder 32B Instruct","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-08-03","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[46]","observedAt":"2025-08-03"}]},{"modelRef":"gpt-oss-120b","rowCount":2,"numericRowCount":2,"benchmarkCount":2,"benchmarkIds":["swebench-bash-only","swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":2},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-08-07","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[0]=bash-only;results[40]","observedAt":"2025-08-07"}]},{"modelRef":"INCModel/Kimi-K2.6-MXFP4-CT-AutoRound","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6","sourceLocator":"rank=5;row=4"}]},{"modelRef":"LGAI-EXAONE/EXAONE-4.5-33B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://huggingface.co/LGAI-EXAONE/EXAONE-4.5-33B","sourceLocator":"rank=18;row=17"}]},{"modelRef":"Qwen/Qwen3-30B-A3B-Thinking-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://matharena.ai/?comp=aime--aime_2026","sourceLocator":"rank=21;row=20"}]},{"modelRef":"Qwen/Qwen3-4B-Thinking-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://matharena.ai/?comp=aime--aime_2026","sourceLocator":"rank=23;row=22"}]},{"modelRef":"inclusionAI/Ling-2.6-flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://huggingface.co/inclusionAI/Ling-2.6-flash","sourceLocator":"rank=24;row=23"}]},{"modelRef":"inclusionAI/Ling-3.0-flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://huggingface.co/inclusionAI/Ling-3.0-flash","sourceLocator":"rank=16;row=15"}]},{"modelRef":"inclusionAI/Ring-2.6-1T","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://huggingface.co/inclusionAI/Ring-2.6-1T","sourceLocator":"rank=8;row=7"}]},{"modelRef":"meta-models/Muse-Glimmer-30B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://huggingface.co/meta-models/Muse-Glimmer-30B","sourceLocator":"rank=12;row=11"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://matharena.ai/?comp=aime--aime_2026","sourceLocator":"rank=20;row=19"}]},{"modelRef":"skt/A.X-K2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://huggingface.co/skt/A.X-K2/blob/fb412dcd22e9c5ac43b2967d6c5e8a7d0e667b0f/README.md#L134-L152","sourceLocator":"rank=2;row=1"}]},{"modelRef":"stepfun-ai/Step-3.5-Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://matharena.ai/?comp=aime--aime_2026","sourceLocator":"rank=3;row=2"}]},{"modelRef":"thinkingmachines/Inkling","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://huggingface.co/thinkingmachines/Inkling","sourceLocator":"rank=1;row=0"}]},{"modelRef":"thinkingmachines/Inkling-Small","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://huggingface.co/thinkingmachines/Inkling-Small","sourceLocator":"rank=10;row=9"}]},{"modelRef":"upstage/Solar-Open2-250B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aime-2026"],"sourceIds":["hf-aime-2026"],"sourceLabels":["Hugging Face · AIME leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-aime-2026","sourceLabel":"Hugging Face · AIME leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/aime_2026/leaderboard","evidenceUrl":"https://huggingface.co/upstage/Solar-Open2-250B","sourceLocator":"rank=9;row=8"}]},{"modelRef":"CohereLabs/c4ai-command-a-03-2025","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/b2/77/b27740dc-a017-4738-8dd0-c18973b61d9e.json","sourceLocator":"rank=102;row=101"}]},{"modelRef":"CohereLabs/c4ai-command-r-08-2024","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/9a/17/9a17fcf2-ee22-4a4f-b992-0d07290a56fc.json","sourceLocator":"rank=113;row=112"}]},{"modelRef":"CohereLabs/c4ai-command-r-plus-08-2024","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/af/03/af03cc06-a493-48b7-a049-61d9b690c8f3.json","sourceLocator":"rank=109;row=108"}]},{"modelRef":"CohereLabs/c4ai-command-r7b-12-2024","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/69/1d/691d4ef0-94b0-47e2-a7c9-b85f1bdc625b.json","sourceLocator":"rank=114;row=113"}]},{"modelRef":"FINAL-Bench/Darwin-27B-Opus","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/FINAL-Bench/Darwin-27B-Opus","sourceLocator":"rank=28;row=27"}]},{"modelRef":"FINAL-Bench/Darwin-28B-REASON","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/FINAL-Bench/Darwin-28B-REASON","sourceLocator":"rank=12;row=11"}]},{"modelRef":"FINAL-Bench/Darwin-31B-Opus","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/FINAL-Bench/Darwin-31B-Opus","sourceLocator":"rank=36;row=35"}]},{"modelRef":"FINAL-Bench/Darwin-36B-Opus","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/FINAL-Bench/Darwin-36B-Opus","sourceLocator":"rank=16;row=15"}]},{"modelRef":"FINAL-Bench/Darwin-398B-JGOS","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/FINAL-Bench/Darwin-398B-JGOS","sourceLocator":"rank=6;row=5"}]},{"modelRef":"FINAL-Bench/Darwin-4B-David","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/FINAL-Bench/Darwin-4B-David","sourceLocator":"rank=42;row=41"}]},{"modelRef":"FINAL-Bench/Darwin-60B-DUO","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/FINAL-Bench/Darwin-60B-DUO","sourceLocator":"rank=17;row=16"}]},{"modelRef":"FINAL-Bench/Darwin-9B-NEG","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/FINAL-Bench/Darwin-9B-NEG","sourceLocator":"rank=45;row=44"}]},{"modelRef":"FINAL-Bench/Ourbox-35B-JGOS","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/FINAL-Bench/Ourbox-35B-JGOS","sourceLocator":"rank=31;row=30"}]},{"modelRef":"FlagRelease/Darwin-9B-NEG-FINAL-hygon-FlagOS","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/FINAL-Bench/Darwin-9B-NEG","sourceLocator":"rank=47;row=46"}]},{"modelRef":"FlagRelease/Darwin-9B-NEG-ansulev-hygon-FlagOS","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/FINAL-Bench/Darwin-9B-NEG","sourceLocator":"rank=48;row=47"}]},{"modelRef":"INCModel/Kimi-K2.6-MXFP4-CT-AutoRound","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6","sourceLocator":"rank=8;row=7"}]},{"modelRef":"JGOS-Model/JGOS-31B-Citizen","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/JGOS-Model/JGOS-31B-Citizen","sourceLocator":"rank=46;row=45"}]},{"modelRef":"JetBrains/Mellum2-12B-A2.5B-Base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=112;row=111"}]},{"modelRef":"JetBrains/Mellum2-12B-A2.5B-Base-Pretrain","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=111;row=110"}]},{"modelRef":"JetBrains/Mellum2-12B-A2.5B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=106;row=105"}]},{"modelRef":"JetBrains/Mellum2-12B-A2.5B-Instruct-SFT","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=108;row=107"}]},{"modelRef":"JetBrains/Mellum2-12B-A2.5B-Thinking","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=100;row=99"}]},{"modelRef":"JetBrains/Mellum2-12B-A2.5B-Thinking-SFT","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","sourceLocator":"rank=107;row=106"}]},{"modelRef":"LGAI-EXAONE/EXAONE-4.5-33B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/LGAI-EXAONE/EXAONE-4.5-33B","sourceLocator":"rank=65;row=64"}]},{"modelRef":"LGAI-EXAONE/K-EXAONE-236B-A23B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/LGAI-EXAONE/K-EXAONE-236B-A23B","sourceLocator":"rank=73;row=72"}]},{"modelRef":"Qwen/Qwen3-30B-A3B-Thinking-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/4b/02/4b02c12c-3f7b-45c8-b746-c590860a7a36.json","sourceLocator":"rank=85;row=84"}]},{"modelRef":"Qwen/Qwen3-4B-Instruct-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507","sourceLocator":"rank=96;row=95"}]},{"modelRef":"Qwen/Qwen3-4B-Thinking-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/Qwen/Qwen3-4B-Thinking-2507","sourceLocator":"rank=95;row=94"}]},{"modelRef":"deepseek-ai/DeepSeek-R1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-R1","sourceLocator":"rank=87;row=86"}]},{"modelRef":"deepseek-ai/DeepSeek-V3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/0a/db/0adbcd9f-38e6-414a-88b1-7bd928d4c72e.json","sourceLocator":"rank=98;row=97"}]},{"modelRef":"ibm-granite/granite-4.1-30b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/ibm-granite/granite-4.1-30b","sourceLocator":"rank=104;row=103"}]},{"modelRef":"ibm-granite/granite-4.1-3b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/ibm-granite/granite-4.1-3b","sourceLocator":"rank=110;row=109"}]},{"modelRef":"ibm-granite/granite-4.1-8b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/ibm-granite/granite-4.1-8b","sourceLocator":"rank=105;row=104"}]},{"modelRef":"inclusionAI/Ring-2.6-1T","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/inclusionAI/Ring-2.6-1T","sourceLocator":"rank=18;row=17"}]},{"modelRef":"jdopensource/JoyAI-LLM-Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/jdopensource/JoyAI-LLM-Flash","sourceLocator":"rank=80;row=79"}]},{"modelRef":"meituan-longcat/LongCat-Flash-Thinking-2601","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/meituan-longcat/LongCat-Flash-Thinking-2601","sourceLocator":"rank=64;row=63"}]},{"modelRef":"meta-llama/Llama-3.2-1B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/aa/ad/aaad0db3-7c5b-4834-a1a2-d96624bdd003.json","sourceLocator":"rank=115;row=114"}]},{"modelRef":"meta-llama/Llama-3.2-90B-Vision-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/94/0f/940ff5c0-b932-4c37-9eb5-27cbd4997484.json","sourceLocator":"rank=103;row=102"}]},{"modelRef":"meta-models/Muse-Glimmer-30B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/meta-models/Muse-Glimmer-30B","sourceLocator":"rank=51;row=50"}]},{"modelRef":"mistralai/Mistral-Small-4-119B-2603","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/mistralai/Mistral-Small-4-119B-2603","sourceLocator":"rank=89;row=88"}]},{"modelRef":"moonshotai/Kimi-K2-Thinking","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2-Thinking","sourceLocator":"rank=44;row=43"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-FP8","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/bd/1f/bd1fe04f-7d58-4045-be60-7150500d537f.json","sourceLocator":"rank=75;row=74"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16","sourceLocator":"rank=27;row=26"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4","sourceLocator":"rank=20;row=19"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16","sourceLocator":"rank=78;row=77"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4","sourceLocator":"rank=77;row=76"}]},{"modelRef":"nvidia/Nemotron-Cascade-2-30B-A3B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/Nemotron-Cascade-2-30B-A3B","sourceLocator":"rank=76;row=75"}]},{"modelRef":"ornith-ai/Ornith-1.5-35B-A3B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B","sourceLocator":"rank=14;row=13"}]},{"modelRef":"ornith-ai/Ornith-1.5-397B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-397B","sourceLocator":"rank=2;row=1"}]},{"modelRef":"ornith-ai/Ornith-1.5-9B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-9B","sourceLocator":"rank=30;row=29"}]},{"modelRef":"skt/A.X-K1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/skt/A.X-K1","sourceLocator":"rank=82;row=81"}]},{"modelRef":"skt/A.X-K2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/skt/A.X-K2/blob/fb412dcd22e9c5ac43b2967d6c5e8a7d0e667b0f/README.md#L134-L152","sourceLocator":"rank=38;row=37"}]},{"modelRef":"stepfun-ai/Step-3.5-Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://arxiv.org/abs/2602.10604","sourceLocator":"rank=50;row=49"}]},{"modelRef":"tencent/Hy3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/tencent/Hy3","sourceLocator":"rank=9;row=8"}]},{"modelRef":"tencent/Hy3-preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/tencent/Hy3-preview","sourceLocator":"rank=25;row=24"}]},{"modelRef":"thinkingmachines/Inkling","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/thinkingmachines/Inkling","sourceLocator":"rank=26;row=25"}]},{"modelRef":"thinkingmachines/Inkling-Small","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/thinkingmachines/Inkling-Small","sourceLocator":"rank=11;row=10"}]},{"modelRef":"upstage/Solar-Open2-250B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/upstage/Solar-Open2-250B","sourceLocator":"rank=32;row=31"}]},{"modelRef":"zai-org/GLM-4.5-Air","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/9c/77/9c7740fd-9cad-445c-9681-eb576e2a110e.json","sourceLocator":"rank=86;row=85"}]},{"modelRef":"zai-org/GLM-4.6-FP8","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["hf-gpqa"],"sourceLabels":["Hugging Face · GPQA leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-gpqa","sourceLabel":"Hugging Face · GPQA leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/Idavidrein/gpqa/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/7e/18/7e18ad63-4a0a-4d9d-a4ec-40b683b42981.json","sourceLocator":"rank=60;row=59"}]},{"modelRef":"HelpingAI/Dhanishtha-2.0-0126","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/HelpingAI/Dhanishtha-2.0-0126","sourceLocator":"rank=80;row=79"}]},{"modelRef":"InternScience/Agents-A1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/InternScience/Agents-A1","sourceLocator":"rank=16;row=15"}]},{"modelRef":"LGAI-EXAONE/K-EXAONE-236B-A23B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/LGAI-EXAONE/K-EXAONE-236B-A23B","sourceLocator":"rank=74;row=73"}]},{"modelRef":"MiniMaxAI/MiniMax-M2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M2","sourceLocator":"rank=75;row=74"}]},{"modelRef":"PolarSeeker/OpenSeeker-v2-30B-SFT","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/PolarSeeker/OpenSeeker-v2-30B-SFT","sourceLocator":"rank=31;row=30"}]},{"modelRef":"dots-studio/dots3-note-prev","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/dots-studio/dots3-note-prev","sourceLocator":"rank=8;row=7"}]},{"modelRef":"inclusionAI/Ling-3.0-flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/inclusionAI/Ling-3.0-flash","sourceLocator":"rank=57;row=56"}]},{"modelRef":"internlm/Intern-S2-Preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/internlm/Intern-S2-Preview","sourceLocator":"rank=61;row=60"}]},{"modelRef":"meta-models/Muse-Glimmer-30B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/meta-models/Muse-Glimmer-30B","sourceLocator":"rank=60;row=59"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16","sourceLocator":"rank=76;row=75"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4","sourceLocator":"rank=79;row=78"}]},{"modelRef":"skt/A.X-K1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/skt/A.X-K1","sourceLocator":"rank=84;row=83"}]},{"modelRef":"skt/A.X-K2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/skt/A.X-K2/blob/fb412dcd22e9c5ac43b2967d6c5e8a7d0e667b0f/README.md#L134-L152","sourceLocator":"rank=41;row=40"}]},{"modelRef":"stepfun-ai/Step-3.5-Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://arxiv.org/abs/2602.10604","sourceLocator":"rank=51;row=50"}]},{"modelRef":"stepfun-ai/Step-3.7-Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/stepfun-ai/Step-3.7-Flash","sourceLocator":"rank=14;row=13"}]},{"modelRef":"tencent/Hy3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/tencent/Hy3","sourceLocator":"rank=7;row=6"}]},{"modelRef":"tencent/Hy3-preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/tencent/Hy3-preview","sourceLocator":"rank=38;row=37"}]},{"modelRef":"thinkingmachines/Inkling","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/thinkingmachines/Inkling","sourceLocator":"rank=18;row=17"}]},{"modelRef":"thinkingmachines/Inkling-Small","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/thinkingmachines/Inkling-Small","sourceLocator":"rank=33;row=32"}]},{"modelRef":"upstage/Solar-Open2-250B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/upstage/Solar-Open2-250B","sourceLocator":"rank=39;row=38"}]},{"modelRef":"zai-org/GLM-4.5","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/93/cb/93cbdfdf-9ec6-4afb-87f4-d771cf29a378.json","sourceLocator":"rank=85;row=84"}]},{"modelRef":"zai-org/GLM-4.5-Air","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hle"],"sourceIds":["hf-hle"],"sourceLabels":["Hugging Face · Humanity's Last Exam leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/cais/hle/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hle","sourceLabel":"Hugging Face · Humanity's Last Exam leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/cais/hle/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/e9/03/e90308ab-e898-4434-aa93-13d131887737.json","sourceLocator":"rank=86;row=85"}]},{"modelRef":"INCModel/Kimi-K2.6-MXFP4-CT-AutoRound","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hmmt"],"sourceIds":["hf-hmmt-2026"],"sourceLabels":["Hugging Face · HMMT leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6","sourceLocator":"rank=2;row=1"}]},{"modelRef":"Qwen/Qwen3-30B-A3B-Thinking-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hmmt"],"sourceIds":["hf-hmmt-2026"],"sourceLabels":["Hugging Face · HMMT leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","evidenceUrl":"https://matharena.ai/?comp=hmmt--hmmt_feb_2026","sourceLocator":"rank=15;row=14"}]},{"modelRef":"Qwen/Qwen3-4B-Thinking-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hmmt"],"sourceIds":["hf-hmmt-2026"],"sourceLabels":["Hugging Face · HMMT leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","evidenceUrl":"https://matharena.ai/?comp=hmmt--hmmt_feb_2026","sourceLocator":"rank=16;row=15"}]},{"modelRef":"inclusionAI/Ling-2.6-flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hmmt"],"sourceIds":["hf-hmmt-2026"],"sourceLabels":["Hugging Face · HMMT leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","evidenceUrl":"https://huggingface.co/inclusionAI/Ling-2.6-flash","sourceLocator":"rank=17;row=16"}]},{"modelRef":"inclusionAI/Ling-3.0-flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hmmt"],"sourceIds":["hf-hmmt-2026"],"sourceLabels":["Hugging Face · HMMT leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","evidenceUrl":"https://huggingface.co/inclusionAI/Ling-3.0-flash","sourceLocator":"rank=7;row=6"}]},{"modelRef":"internlm/Intern-S2-Preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hmmt"],"sourceIds":["hf-hmmt-2026"],"sourceLabels":["Hugging Face · HMMT leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","evidenceUrl":"https://huggingface.co/internlm/Intern-S2-Preview","sourceLocator":"rank=5;row=4"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hmmt"],"sourceIds":["hf-hmmt-2026"],"sourceLabels":["Hugging Face · HMMT leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","evidenceUrl":"https://matharena.ai/?comp=hmmt--hmmt_feb_2026","sourceLocator":"rank=10;row=9"}]},{"modelRef":"stepfun-ai/Step-3.5-Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hmmt"],"sourceIds":["hf-hmmt-2026"],"sourceLabels":["Hugging Face · HMMT leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","evidenceUrl":"https://matharena.ai/?comp=hmmt--hmmt_feb_2026","sourceLocator":"rank=8;row=7"}]},{"modelRef":"thinkingmachines/Inkling-Small","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["hmmt"],"sourceIds":["hf-hmmt-2026"],"sourceLabels":["Hugging Face · HMMT leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-hmmt-2026","sourceLabel":"Hugging Face · HMMT leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/MathArena/hmmt_feb_2026/leaderboard","evidenceUrl":"https://huggingface.co/thinkingmachines/Inkling-Small","sourceLocator":"rank=3;row=2"}]},{"modelRef":"01-ai/Yi-1.5-34B-Chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/67/6f/676f4465-ce78-411a-9f5a-c97b3d2eac4f.json","sourceLocator":"rank=80;row=79"}]},{"modelRef":"01-ai/Yi-1.5-6B-Chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/bd/05/bd056a61-8ede-45b4-823d-093343bdd880.json","sourceLocator":"rank=108;row=107"}]},{"modelRef":"01-ai/Yi-1.5-9B-Chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/94/9e/949ec028-f03f-4211-8783-c810af6489a4.json","sourceLocator":"rank=90;row=89"}]},{"modelRef":"01-ai/Yi-34B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/9d/42/9d42931c-6ec3-48ac-ae0a-cebb9c1f68aa.json","sourceLocator":"rank=98;row=97"}]},{"modelRef":"01-ai/Yi-6B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/8b/02/8b02ff6f-b3ae-4230-8edd-50c3ffa34592.json","sourceLocator":"rank=128;row=127"}]},{"modelRef":"01-ai/Yi-6B-Chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/28/1d/281df2d2-60b5-4053-b30e-bba4976c3efd.json","sourceLocator":"rank=127;row=126"}]},{"modelRef":"ByteDance-Seed/Seed-OSS-36B-Base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/cf/2d/cf2d98c9-1e57-4994-9e49-9a7962a2706d.json","sourceLocator":"rank=57;row=56"}]},{"modelRef":"ByteDance-Seed/Seed-OSS-36B-Base-woSyn","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/60/1d/601d782d-660b-4f01-bcae-5c4105223c85.json","sourceLocator":"rank=65;row=64"}]},{"modelRef":"ByteDance-Seed/Seed-OSS-36B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/1b/40/1b405e01-7137-4189-b5d2-83e0f4004dc9.json","sourceLocator":"rank=28;row=27"}]},{"modelRef":"CohereLabs/aya-expanse-32b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/12/29/1229c8e0-d21f-43e8-a9c7-3ea0d09408d5.json","sourceLocator":"rank=92;row=91"}]},{"modelRef":"CohereLabs/aya-expanse-8b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/33/5b/335b15a4-a7fd-4ffc-81db-9d5feed90657.json","sourceLocator":"rank=119;row=118"}]},{"modelRef":"CohereLabs/c4ai-command-r-v01","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/b3/5f/b35fcd95-c193-4ed0-b50c-bc535af5756f.json","sourceLocator":"rank=111;row=110"}]},{"modelRef":"EleutherAI/llemma_7b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/f0/87/f087fc33-d4b9-4939-b3cf-2d3fb9094177.json","sourceLocator":"rank=130;row=129"}]},{"modelRef":"GSAI-ML/LLaDA-8B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/67/cc/67cc35fb-ae5a-4632-91da-9346a6b2849e.json","sourceLocator":"rank=113;row=112"}]},{"modelRef":"HuggingFaceTB/SmolLM2-1.7B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/33/c9/33c9180c-621e-40a2-bb6f-02b90ecff1fd.json","sourceLocator":"rank=136;row=135"}]},{"modelRef":"HuggingFaceTB/SmolLM2-135M","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/ac/90/ac90804d-cb99-41cc-b3e7-65ed31484008.json","sourceLocator":"rank=142;row=141"}]},{"modelRef":"HuggingFaceTB/SmolLM2-360M","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/23/ae/23aee40b-5bea-46cf-a87b-8263e12e2d1f.json","sourceLocator":"rank=141;row=140"}]},{"modelRef":"JetBrains/Mellum2-12B-A2.5B-Base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=68;row=67"}]},{"modelRef":"JetBrains/Mellum2-12B-A2.5B-Base-Pretrain","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","sourceLocator":"rank=67;row=66"}]},{"modelRef":"LGAI-EXAONE/EXAONE-3.5-2.4B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/ce/1a/ce1a8a1a-5d16-4403-9e85-bf0f3d72f09f.json","sourceLocator":"rank=107;row=106"}]},{"modelRef":"LGAI-EXAONE/EXAONE-3.5-32B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/de/c9/dec9e4cf-cc67-42e1-9474-f8de9048014b.json","sourceLocator":"rank=69;row=68"}]},{"modelRef":"LGAI-EXAONE/EXAONE-3.5-7.8B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/c6/8f/c68fc694-d6dd-4b2a-a699-e5f737ae550f.json","sourceLocator":"rank=89;row=88"}]},{"modelRef":"LGAI-EXAONE/EXAONE-4.5-33B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/LGAI-EXAONE/EXAONE-4.5-33B","sourceLocator":"rank=26;row=25"}]},{"modelRef":"LGAI-EXAONE/K-EXAONE-236B-A23B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/LGAI-EXAONE/K-EXAONE-236B-A23B","sourceLocator":"rank=19;row=18"}]},{"modelRef":"MiniMaxAI/MiniMax-M1-40k","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/3b/7b/3b7bfaac-ee52-4508-8ecd-676d82e06c1c.json","sourceLocator":"rank=36;row=35"}]},{"modelRef":"MiniMaxAI/MiniMax-M2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M2","sourceLocator":"rank=30;row=29"}]},{"modelRef":"MiniMaxAI/MiniMax-Text-01","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/2a/73/2a734d0f-7fcf-48d0-bea6-ad17cd749ee4.json","sourceLocator":"rank=44;row=43"}]},{"modelRef":"Qwen/Qwen1.5-110B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/33/3e/333e0d9f-7e39-4da3-96cb-0501d2380c03.json","sourceLocator":"rank=83;row=82"}]},{"modelRef":"Qwen/Qwen1.5-14B-Chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/09/ca/09ca61cf-1158-46e6-aeae-1d68f5dfbc34.json","sourceLocator":"rank=109;row=108"}]},{"modelRef":"Qwen/Qwen1.5-7B-Chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/db/e1/dbe12346-25db-4fac-9421-5153fc4c661c.json","sourceLocator":"rank=126;row=125"}]},{"modelRef":"Qwen/Qwen2-0.5B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/85/72/8572a257-22bc-45c5-8891-d698ba6fc206.json","sourceLocator":"rank=137;row=136"}]},{"modelRef":"Qwen/Qwen2-1.5B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/ff/fc/fffc978a-e6ae-41f2-bded-06bb776ee8b9.json","sourceLocator":"rank=132;row=131"}]},{"modelRef":"Qwen/Qwen2-1.5B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/18/ea/18ea6501-7d93-4ec4-bd82-86991bb5cc68.json","sourceLocator":"rank=131;row=130"}]},{"modelRef":"Qwen/Qwen2-7B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/37/79/3779cfdb-412d-49fa-94d4-5c7ae1af8ac6.json","sourceLocator":"rank=105;row=104"}]},{"modelRef":"Qwen/Qwen2.5-0.5B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/5b/03/5b03aafd-e973-4ea4-850c-2ab97c1abb27.json","sourceLocator":"rank=138;row=137"}]},{"modelRef":"Qwen/Qwen2.5-1.5B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/e1/d5/e1d5921c-130a-44f8-ae14-baed069e565e.json","sourceLocator":"rank=122;row=121"}]},{"modelRef":"Qwen/Qwen2.5-14B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/cc/7c/cc7c2263-9baf-45fc-b58b-64c15989e358.json","sourceLocator":"rank=61;row=60"}]},{"modelRef":"Qwen/Qwen2.5-32B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/da/c5/dac51285-4928-4ac7-b052-510e4705981b.json","sourceLocator":"rank=50;row=49"}]},{"modelRef":"Qwen/Qwen2.5-3B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/ed/3a/ed3af7f3-5881-4ddd-bab1-75ecda64c98e.json","sourceLocator":"rank=96;row=95"}]},{"modelRef":"Qwen/Qwen2.5-72B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/06/60/06600438-ea9f-4078-8eef-701c59af71a4.json","sourceLocator":"rank=46;row=45"}]},{"modelRef":"Qwen/Qwen2.5-7B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/0c/64/0c645bb3-1e9c-4400-a704-f99748356114.json","sourceLocator":"rank=94;row=93"}]},{"modelRef":"Qwen/Qwen2.5-VL-72B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/9c/45/9c455b8c-1153-4c24-9de0-1d506297d805.json","sourceLocator":"rank=143;row=142"}]},{"modelRef":"Qwen/Qwen3-235B-A22B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/99/98/99989790-84de-4fd9-af37-1d9c28dd95b3.json","sourceLocator":"rank=52;row=51"}]},{"modelRef":"Qwen/Qwen3-235B-A22B-Instruct-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/df/98/df9861f7-59cf-4b5e-8988-571481f47d14.json","sourceLocator":"rank=27;row=26"}]},{"modelRef":"Qwen/Qwen3-30B-A3B-Base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/74/e6/74e614fc-75a5-4f12-b077-cbf525141a9f.json","sourceLocator":"rank=62;row=61"}]},{"modelRef":"Qwen/Qwen3-30B-A3B-Thinking-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/46/a3/46a34cab-9475-40a1-ba07-2c0528f6bf5d.json","sourceLocator":"rank=39;row=38"}]},{"modelRef":"Qwen/Qwen3-4B-Instruct-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/Qwen/Qwen3-4B-Instruct-2507","sourceLocator":"rank=48;row=47"}]},{"modelRef":"RedHatAI/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=21;row=20"}]},{"modelRef":"TIGER-Lab/MAmmoTH2-7B-Plus","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/1b/16/1b16feea-1def-49ae-b3da-ee3af193a705.json","sourceLocator":"rank=104;row=103"}]},{"modelRef":"TIGER-Lab/MAmmoTH2-8B-Plus","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/1f/96/1f963bf0-4985-4d1f-a322-55865369de75.json","sourceLocator":"rank=97;row=96"}]},{"modelRef":"TIGER-Lab/MAmmoTH2-8x7B-Plus","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/17/d1/17d1bf78-2fbe-414d-b555-6d672f57ddf2.json","sourceLocator":"rank=82;row=81"}]},{"modelRef":"XiaomiMiMo/MiMo-7B-Base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/67/a9/67a990d8-bf28-4d9e-94b8-a472bde1efbd.json","sourceLocator":"rank=100;row=99"}]},{"modelRef":"XiaomiMiMo/MiMo-7B-RL","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/eb/b6/ebb62699-a109-4160-9f12-152a17ac9c5b.json","sourceLocator":"rank=70;row=69"}]},{"modelRef":"abacusai/Llama-3-Smaug-8B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/b0/a9/b0a990c3-c1a2-4a69-abed-e1c2c51d8763.json","sourceLocator":"rank=114;row=113"}]},{"modelRef":"ai21labs/AI21-Jamba-Large-1.5","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/aa/2b/aa2be737-1a21-4cc9-93f2-83316b7dc0cd.json","sourceLocator":"rank=85;row=84"}]},{"modelRef":"baidu/ERNIE-4.5-21B-A3B-Base-PT","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/5d/0e/5d0edbf1-b369-43eb-b885-d5a87c818d1f.json","sourceLocator":"rank=72;row=71"}]},{"modelRef":"baidu/ERNIE-4.5-300B-A47B-Base-PT","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/bc/f1/bcf18d24-d600-4ed9-ba95-73f8d58569e1.json","sourceLocator":"rank=49;row=48"}]},{"modelRef":"baidu/ERNIE-4.5-300B-A47B-PT","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/e6/6b/e66bffdc-f3d1-4808-90a4-43f204aa883e.json","sourceLocator":"rank=42;row=41"}]},{"modelRef":"deepseek-ai/DeepSeek-Coder-V2-Lite-Base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/1a/c8/1ac85691-ef95-4bf5-8618-6bac456e2792.json","sourceLocator":"rank=118;row=117"}]},{"modelRef":"deepseek-ai/DeepSeek-Coder-V2-Lite-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/0c/0f/0c0f446a-6c2a-47c7-a120-4fafb90f92a3.json","sourceLocator":"rank=101;row=100"}]},{"modelRef":"deepseek-ai/DeepSeek-R1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-R1","sourceLocator":"rank=18;row=17"}]},{"modelRef":"deepseek-ai/DeepSeek-V2-Chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/a7/68/a768afc2-e0b4-4ea5-ba40-166136c33b3b.json","sourceLocator":"rank=75;row=74"}]},{"modelRef":"deepseek-ai/DeepSeek-V2.5","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/65/c2/65c29cb3-af56-45f1-be1b-751630185a59.json","sourceLocator":"rank=56;row=55"}]},{"modelRef":"deepseek-ai/DeepSeek-V3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V3","sourceLocator":"rank=59;row=58"}]},{"modelRef":"deepseek-ai/DeepSeek-V3-0324","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/deepseek-ai/DeepSeek-V3-0324","sourceLocator":"rank=35;row=34"}]},{"modelRef":"deepseek-ai/deepseek-math-7b-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/65/7f/657f147a-6450-42fe-a10d-e626f9000d29.json","sourceLocator":"rank=117;row=116"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-MXFP4_MOE-dequant-bf16-vllm","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=24;row=23"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q3_K_M-dequant-bf16-vllm","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=22;row=21"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_M-dequant-bf16-vllm","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=23;row=22"}]},{"modelRef":"google/gemma-2-9b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/76/6f/766fb522-5c40-49f9-bcad-bd5496404129.json","sourceLocator":"rank=93;row=92"}]},{"modelRef":"ibm-granite/granite-3.0-2b-base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/06/7f/067fe2c1-7ba3-4f79-8344-e55e999ae537.json","sourceLocator":"rank=134;row=133"}]},{"modelRef":"ibm-granite/granite-3.0-8b-base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/d9/f9/d9f9b73a-5b6d-418b-8f26-f89b4d2b6792.json","sourceLocator":"rank=123;row=122"}]},{"modelRef":"ibm-granite/granite-3.1-1b-a400m-base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/4a/f2/4af2fd82-845a-4087-83eb-ae9b4b7561d1.json","sourceLocator":"rank=139;row=138"}]},{"modelRef":"ibm-granite/granite-3.1-2b-base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/d4/c0/d4c06d3d-8a53-44e5-9db6-ebffc99390d2.json","sourceLocator":"rank=129;row=128"}]},{"modelRef":"ibm-granite/granite-3.1-3b-a800m-base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/a9/31/a931868f-9ae1-4cd5-a2a7-8ea402101bbe.json","sourceLocator":"rank=135;row=134"}]},{"modelRef":"ibm-granite/granite-3.1-8b-base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/5b/c9/5bc90039-eb1c-4772-aa34-1cfb7c2343b6.json","sourceLocator":"rank=121;row=120"}]},{"modelRef":"ibm-granite/granite-4.1-30b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/ibm-granite/granite-4.1-30b","sourceLocator":"rank=60;row=59"}]},{"modelRef":"ibm-granite/granite-4.1-3b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/ibm-granite/granite-4.1-3b","sourceLocator":"rank=84;row=83"}]},{"modelRef":"ibm-granite/granite-4.1-8b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/ibm-granite/granite-4.1-8b","sourceLocator":"rank=73;row=72"}]},{"modelRef":"internlm/Intern-S1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/4a/0a/4a0a0ab9-bb59-411e-a406-f6ac49a1e9f1.json","sourceLocator":"rank=25;row=24"}]},{"modelRef":"internlm/Intern-S2-Preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/internlm/Intern-S2-Preview","sourceLocator":"rank=2;row=1"}]},{"modelRef":"internlm/internlm2-math-plus-20b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/6f/ee/6fee4d03-47d0-4c2e-b449-978442fa002c.json","sourceLocator":"rank=112;row=111"}]},{"modelRef":"internlm/internlm2-math-plus-7b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/37/20/37203543-7ca5-4441-be5c-e4cbd12293c2.json","sourceLocator":"rank=120;row=119"}]},{"modelRef":"internlm/internlm3-8b-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/46/a7/46a7e0d0-c351-46ac-99f0-98aebcf3fef7.json","sourceLocator":"rank=71;row=70"}]},{"modelRef":"jdopensource/JoyAI-LLM-Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/jdopensource/JoyAI-LLM-Flash","sourceLocator":"rank=37;row=36"}]},{"modelRef":"meituan-longcat/LongCat-Flash-Chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/19/91/1991ed80-048e-428c-a9f7-e32c71bd65ea.json","sourceLocator":"rank=29;row=28"}]},{"modelRef":"meituan-longcat/LongCat-Flash-Lite","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/meituan-longcat/LongCat-Flash-Lite","sourceLocator":"rank=43;row=42"}]},{"modelRef":"meta-llama/Llama-3.1-405B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/50/a3/50a33ec6-cf9b-4acc-9f45-61cd8356207c.json","sourceLocator":"rank=63;row=62"}]},{"modelRef":"meta-llama/Llama-3.1-70B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/3b/18/3b18649e-fd29-4193-a8c1-6d6d93d1ebd2.json","sourceLocator":"rank=79;row=78"}]},{"modelRef":"meta-llama/Llama-3.1-8B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/45/ba/45baf8ce-a6cc-412f-a134-2242a4af4c93.json","sourceLocator":"rank=115;row=114"}]},{"modelRef":"meta-llama/Llama-3.2-1B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/d5/a3/d5a3c88a-3fee-40b3-a8c7-8c4cc65baf5f.json","sourceLocator":"rank=140;row=139"}]},{"modelRef":"meta-llama/Llama-3.2-3B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/91/89/9189a395-7ef2-453e-84e0-259870efc2f7.json","sourceLocator":"rank=133;row=132"}]},{"modelRef":"meta-llama/Meta-Llama-3-70B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/77/50/77508bc1-dc52-4a84-9df1-deb62975c053.json","sourceLocator":"rank=78;row=77"}]},{"modelRef":"meta-llama/Meta-Llama-3-8B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/ac/75/ac75fe42-62e5-4ea9-a130-1eeacc8e9fb2.json","sourceLocator":"rank=116;row=115"}]},{"modelRef":"meta-llama/Meta-Llama-3-8B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/3c/1f/3c1f9257-27aa-4da0-b772-60d62242a72b.json","sourceLocator":"rank=103;row=102"}]},{"modelRef":"microsoft/Phi-3-medium-128k-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/22/1e/221e4697-1155-4d8d-87df-9778573a3ad7.json","sourceLocator":"rank=81;row=80"}]},{"modelRef":"microsoft/Phi-3-medium-4k-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/63/f9/63f9e7e6-d69f-4522-a81d-7280eacf730c.json","sourceLocator":"rank=74;row=73"}]},{"modelRef":"microsoft/Phi-3-mini-128k-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/ab/4d/ab4d706d-1629-46ac-8db3-cb6fac6de332.json","sourceLocator":"rank=95;row=94"}]},{"modelRef":"microsoft/Phi-3-mini-4k-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/05/3a/053a7b11-0d89-4289-bfe4-ed3eca7c4a97.json","sourceLocator":"rank=91;row=90"}]},{"modelRef":"microsoft/Phi-3.5-mini-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/03/b7/03b77a5a-1723-4cab-9d50-1a634bf8b06a.json","sourceLocator":"rank=88;row=87"}]},{"modelRef":"microsoft/Phi-4-mini-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/b2/6e/b26e73bc-2a30-4fa4-8223-148df3d82658.json","sourceLocator":"rank=77;row=76"}]},{"modelRef":"microsoft/phi-4","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/c2/37/c23724de-5d83-4794-abcb-9af3ce249886.json","sourceLocator":"rank=47;row=46"}]},{"modelRef":"mistral-community/Mistral-7B-v0.2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/93/21/9321cec5-309d-4328-aa2c-709af17cff9c.json","sourceLocator":"rank=125;row=124"}]},{"modelRef":"mistralai/Mathstral-7B-v0.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/b7/6c/b76c036b-b67b-4dd4-9c7f-d1938b83f254.json","sourceLocator":"rank=99;row=98"}]},{"modelRef":"mistralai/Ministral-8B-Instruct-2410","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/b0/f5/b0f59e78-883f-4c57-b728-b983c859f23f.json","sourceLocator":"rank=110;row=109"}]},{"modelRef":"mistralai/Mistral-7B-Instruct-v0.2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/9d/8d/9d8dc94b-067c-478d-aa99-874b71721fc3.json","sourceLocator":"rank=124;row=123"}]},{"modelRef":"mistralai/Mistral-Large-Instruct-2407","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/b7/99/b799f854-5a9c-413c-b766-ab127e0cd103.json","sourceLocator":"rank=55;row=54"}]},{"modelRef":"mistralai/Mistral-Large-Instruct-2411","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/03/98/03985ef7-288d-4421-9257-ce6b77dcd858.json","sourceLocator":"rank=53;row=52"}]},{"modelRef":"mistralai/Mistral-Nemo-Base-2407","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/be/91/be919504-16fb-4669-8d9b-3bfdfd2e6364.json","sourceLocator":"rank=106;row=105"}]},{"modelRef":"mistralai/Mistral-Small-24B-Base-2501","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/d4/89/d489ea6c-1287-43bd-aa64-5193c09331b2.json","sourceLocator":"rank=76;row=75"}]},{"modelRef":"mistralai/Mistral-Small-Instruct-2409","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/71/31/7131c1f0-8fce-4050-8117-51d9b5e651ca.json","sourceLocator":"rank=86;row=85"}]},{"modelRef":"mistralai/Mixtral-8x7B-v0.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/2b/dd/2bddced4-34a8-4a1c-b582-90bfe03346f1.json","sourceLocator":"rank=102;row=101"}]},{"modelRef":"moonshotai/Kimi-K2-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/4a/ae/4aae2732-e24b-42a2-9dc1-a35cd215a8ec.json","sourceLocator":"rank=38;row=37"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/2a/e5/2ae5bb55-f5e3-4aa2-beef-141f450a4562.json","sourceLocator":"rank=58;row=57"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=20;row=19"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16","sourceLocator":"rank=6;row=5"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4","sourceLocator":"rank=7;row=6"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16","sourceLocator":"rank=31;row=30"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4","sourceLocator":"rank=32;row=31"}]},{"modelRef":"nvidia/Nemotron-H-56B-Base-8K","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/8b/b3/8bb31817-e160-4285-a781-272851ae2c41.json","sourceLocator":"rank=64;row=63"}]},{"modelRef":"openai/gpt-oss-120b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/e4/8d/e48dbb1a-9e1a-4c99-abbf-cb32942b48e7.json","sourceLocator":"rank=40;row=39"}]},{"modelRef":"openai/gpt-oss-20b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/ca/ef/caef6dbc-0f07-4491-9009-2b0d26dc9724.json","sourceLocator":"rank=45;row=44"}]},{"modelRef":"skt/A.X-K1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/skt/A.X-K1","sourceLocator":"rank=33;row=32"}]},{"modelRef":"stepfun-ai/Step-3.5-Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://arxiv.org/abs/2602.10604","sourceLocator":"rank=17;row=16"}]},{"modelRef":"tencent/Hunyuan-A13B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/4b/ff/4bffdd42-60c3-41a9-9ef8-e5849e114d9e.json","sourceLocator":"rank=54;row=53"}]},{"modelRef":"tencent/Tencent-Hunyuan-Large","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/5c/9f/5c9f18ed-3247-4af2-b44f-2a504735554b.json","sourceLocator":"rank=66;row=65"}]},{"modelRef":"upstage/Solar-Open2-250B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/upstage/Solar-Open2-250B","sourceLocator":"rank=11;row=10"}]},{"modelRef":"zai-org/GLM-4.5","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/04/c7/04c77f44-dff1-4cc7-80a1-35f4e9e9237f.json","sourceLocator":"rank=16;row=15"}]},{"modelRef":"zai-org/GLM-4.5-Air","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/07/43/074397da-f5e5-441c-bc45-ac16b599ca7c.json","sourceLocator":"rank=34;row=33"}]},{"modelRef":"zai-org/glm-4-9b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["mmlu-pro"],"sourceIds":["hf-mmlu-pro"],"sourceLabels":["Hugging Face · MMLU-Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-mmlu-pro","sourceLabel":"Hugging Face · MMLU-Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/TIGER-Lab/MMLU-Pro/leaderboard","evidenceUrl":"https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/92/ff/92ff34d6-a9d7-48af-a31e-a72c2367edff.json","sourceLocator":"rank=87;row=86"}]},{"modelRef":"CohereLabs/North-Mini-Code-1.0","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/CohereLabs/North-Mini-Code-1.0","sourceLocator":"rank=36;row=35"}]},{"modelRef":"INCModel/Kimi-K2.6-MXFP4-CT-AutoRound","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6","sourceLocator":"rank=12;row=11"}]},{"modelRef":"MuVeraAI/Laguna-XS.2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-XS.2","sourceLocator":"rank=33;row=32"}]},{"modelRef":"Qwen/Qwen3-235B-A22B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://scale.com/leaderboard/swe_bench_pro_public","sourceLocator":"rank=40;row=39"}]},{"modelRef":"dots-studio/dots3-note-prev","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/dots-studio/dots3-note-prev","sourceLocator":"rank=7;row=6"}]},{"modelRef":"inclusionAI/Ling-3.0-flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/inclusionAI/Ling-3.0-flash","sourceLocator":"rank=16;row=15"}]},{"modelRef":"meta-models/Muse-Glimmer-30B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/meta-models/Muse-Glimmer-30B","sourceLocator":"rank=25;row=24"}]},{"modelRef":"moonshotai/Kimi-K2-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://scale.com/leaderboard/swe_bench_pro_public","sourceLocator":"rank=39;row=38"}]},{"modelRef":"openai/gpt-oss-120b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://scale.com/leaderboard/swe_bench_pro_public","sourceLocator":"rank=41;row=40"}]},{"modelRef":"ornith-ai/Ornith-1.0-35B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B","sourceLocator":"rank=27;row=26"}]},{"modelRef":"ornith-ai/Ornith-1.0-397B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B","sourceLocator":"rank=4;row=3"}]},{"modelRef":"ornith-ai/Ornith-1.0-9B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B","sourceLocator":"rank=35;row=34"}]},{"modelRef":"ornith-ai/Ornith-1.5-35B-A3B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B","sourceLocator":"rank=8;row=7"}]},{"modelRef":"ornith-ai/Ornith-1.5-397B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-397B","sourceLocator":"rank=2;row=1"}]},{"modelRef":"ornith-ai/Ornith-1.5-9B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-9B","sourceLocator":"rank=31;row=30"}]},{"modelRef":"poolside/Laguna-M.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-M.1","sourceLocator":"rank=29;row=28"}]},{"modelRef":"poolside/Laguna-S-2.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-S-2.1","sourceLocator":"rank=9;row=8"}]},{"modelRef":"poolside/Laguna-XS-2.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-XS-2.1","sourceLocator":"rank=30;row=29"}]},{"modelRef":"poolside/Laguna-XS.2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-XS.2","sourceLocator":"rank=32;row=31"}]},{"modelRef":"stepfun-ai/Step-3.7-Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/stepfun-ai/Step-3.7-Flash","sourceLocator":"rank=17;row=16"}]},{"modelRef":"tencent/Hy3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/tencent/Hy3","sourceLocator":"rank=14;row=13"}]},{"modelRef":"thinkingmachines/Inkling","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/thinkingmachines/Inkling","sourceLocator":"rank=23;row=22"}]},{"modelRef":"thinkingmachines/Inkling-Small","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://huggingface.co/thinkingmachines/Inkling-Small","sourceLocator":"rank=20;row=19"}]},{"modelRef":"zai-org/GLM-4.6","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-pro"],"sourceIds":["hf-swebench-pro"],"sourceLabels":["Hugging Face · SWE-bench Pro leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-pro","sourceLabel":"Hugging Face · SWE-bench Pro leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/ScaleAI/SWE-bench_Pro/leaderboard","evidenceUrl":"https://scale.com/leaderboard/swe_bench_pro_public","sourceLocator":"rank=42;row=41"}]},{"modelRef":"CohereLabs/North-Mini-Code-1.0","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/CohereLabs/North-Mini-Code-1.0","sourceLocator":"rank=48;row=47"}]},{"modelRef":"INCModel/Kimi-K2.6-MXFP4-CT-AutoRound","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6","sourceLocator":"rank=9;row=8"}]},{"modelRef":"MiniMaxAI/MiniMax-M2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/MiniMaxAI/MiniMax-M2","sourceLocator":"rank=45;row=44"}]},{"modelRef":"MuVeraAI/Laguna-XS.2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-XS.2","sourceLocator":"rank=47;row=46"}]},{"modelRef":"Multilingual-Multimodal-NLP/IndustrialCoder","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/papers/2603.16790","sourceLocator":"rank=24;row=23"}]},{"modelRef":"dots-studio/dots3-note-prev","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/dots-studio/dots3-note-prev","sourceLocator":"rank=13;row=12"}]},{"modelRef":"facebook/cwm","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/papers/2510.02387","sourceLocator":"rank=66;row=65"}]},{"modelRef":"inclusionAI/Ling-2.6-1T","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/inclusionAI/Ling-2.6-1T","sourceLocator":"rank=34;row=33"}]},{"modelRef":"inclusionAI/Ling-2.6-flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/inclusionAI/Ling-2.6-flash","sourceLocator":"rank=52;row=51"}]},{"modelRef":"inclusionAI/Ring-2.6-1T","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/inclusionAI/Ring-2.6-1T","sourceLocator":"rank=29;row=28"}]},{"modelRef":"internlm/Intern-S2-Preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/internlm/Intern-S2-Preview","sourceLocator":"rank=49;row=48"}]},{"modelRef":"meta-models/Muse-Glimmer-30B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/meta-models/Muse-Glimmer-30B","sourceLocator":"rank=20;row=19"}]},{"modelRef":"mindlab-research/Macaron-V1-Coding-Venti","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/mindlab-research/Macaron-V1-Coding-Venti/blob/0a7ec8fdf72e0757decabd90cc3305bed1fed1c6/README.md#L68-L76","sourceLocator":"rank=3;row=2"}]},{"modelRef":"mindlab-research/Macaron-V1-Tall","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/mindlab-research/Macaron-V1-Tall/blob/d0b2199c3572d336bcfe9e027ec519314cf608bd/assets/v1_benchmark.png","sourceLocator":"rank=23;row=22"}]},{"modelRef":"mindlab-research/Macaron-V1-Venti","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/mindlab-research/Macaron-V1-Venti","sourceLocator":"rank=2;row=1"}]},{"modelRef":"mistralai/Mistral-Medium-3.5-128B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/mistralai/Mistral-Medium-3.5-128B","sourceLocator":"rank=16;row=15"}]},{"modelRef":"moonshotai/Kimi-K2-Thinking","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2-Thinking","sourceLocator":"rank=37;row=36"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16","sourceLocator":"rank=36;row=35"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-NVFP4","sourceLocator":"rank=44;row=43"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16","sourceLocator":"rank=75;row=74"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4","sourceLocator":"rank=73;row=72"}]},{"modelRef":"ornith-ai/Ornith-1.0-35B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-35B","sourceLocator":"rank=22;row=21"}]},{"modelRef":"ornith-ai/Ornith-1.0-397B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-397B","sourceLocator":"rank=4;row=3"}]},{"modelRef":"ornith-ai/Ornith-1.0-9B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/deepreinforce-ai/Ornith-1.0-9B","sourceLocator":"rank=46;row=45"}]},{"modelRef":"ornith-ai/Ornith-1.5-35B-A3B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B","sourceLocator":"rank=11;row=10"}]},{"modelRef":"ornith-ai/Ornith-1.5-397B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-397B","sourceLocator":"rank=1;row=0"}]},{"modelRef":"ornith-ai/Ornith-1.5-9B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/ornith-ai/Ornith-1.5-9B","sourceLocator":"rank=41;row=40"}]},{"modelRef":"poolside/Laguna-M.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-M.1","sourceLocator":"rank=25;row=24"}]},{"modelRef":"poolside/Laguna-XS-2.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-XS-2.1","sourceLocator":"rank=38;row=37"}]},{"modelRef":"poolside/Laguna-XS.2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-XS.2","sourceLocator":"rank=43;row=42"}]},{"modelRef":"stepfun-ai/Step-3.5-Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://arxiv.org/abs/2602.10604","sourceLocator":"rank=26;row=25"}]},{"modelRef":"tencent/Hy3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/tencent/Hy3","sourceLocator":"rank=14;row=13"}]},{"modelRef":"tencent/Hy3-preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/tencent/Hy3-preview","sourceLocator":"rank=27;row=26"}]},{"modelRef":"thinkingmachines/Inkling","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/thinkingmachines/Inkling","sourceLocator":"rank=17;row=16"}]},{"modelRef":"thinkingmachines/Inkling-Small","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/thinkingmachines/Inkling-Small","sourceLocator":"rank=8;row=7"}]},{"modelRef":"upstage/Solar-Open2-250B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["hf-swebench-verified"],"sourceLabels":["Hugging Face · SWE-bench Verified leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-swebench-verified","sourceLabel":"Hugging Face · SWE-bench Verified leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/SWE-bench/SWE-bench_Verified/leaderboard","evidenceUrl":"https://huggingface.co/upstage/Solar-Open2-250B","sourceLocator":"rank=42;row=41"}]},{"modelRef":"CohereLabs/North-Mini-Code-1.0","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/CohereLabs/North-Mini-Code-1.0","sourceLocator":"rank=23;row=22"}]},{"modelRef":"INCModel/Kimi-K2.6-MXFP4-CT-AutoRound","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/moonshotai/Kimi-K2.6","sourceLocator":"rank=6;row=5"}]},{"modelRef":"MiniMaxAI/MiniMax-M2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0","sourceLocator":"rank=33;row=32"}]},{"modelRef":"MuVeraAI/Laguna-XS.2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-XS.2","sourceLocator":"rank=32;row=31"}]},{"modelRef":"RedHatAI/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=28;row=27"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-MXFP4_MOE-dequant-bf16-vllm","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=31;row=30"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q3_K_M-dequant-bf16-vllm","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=29;row=28"}]},{"modelRef":"exolabs/NVIDIA-Nemotron-3-Super-120B-A12B-UD-Q4_K_M-dequant-bf16-vllm","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=30;row=29"}]},{"modelRef":"mindlab-research/Macaron-V1-Venti","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/mindlab-research/Macaron-V1-Venti/blob/main/README.md#evaluation","sourceLocator":"rank=1;row=0"}]},{"modelRef":"moonshotai/Kimi-K2-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0","sourceLocator":"rank=35;row=34"}]},{"modelRef":"moonshotai/Kimi-K2-Thinking","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0","sourceLocator":"rank=24;row=23"}]},{"modelRef":"nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16","sourceLocator":"rank=27;row=26"}]},{"modelRef":"nvidia/Nemotron-Terminal-14B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://arxiv.org/pdf/2602.21193","sourceLocator":"rank=39;row=38"}]},{"modelRef":"nvidia/Nemotron-Terminal-32B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://arxiv.org/pdf/2602.21193","sourceLocator":"rank=36;row=35"}]},{"modelRef":"nvidia/Nemotron-Terminal-8B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://arxiv.org/pdf/2602.21193","sourceLocator":"rank=40;row=39"}]},{"modelRef":"poolside/Laguna-M.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-M.1","sourceLocator":"rank=18;row=17"}]},{"modelRef":"poolside/Laguna-XS-2.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-XS-2.1","sourceLocator":"rank=21;row=20"}]},{"modelRef":"poolside/Laguna-XS.2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/poolside/Laguna-XS.2","sourceLocator":"rank=25;row=24"}]},{"modelRef":"stepfun-ai/Step-3.5-Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://arxiv.org/abs/2602.10604","sourceLocator":"rank=16;row=15"}]},{"modelRef":"tencent/Hy3-preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://huggingface.co/tencent/Hy3-preview","sourceLocator":"rank=12;row=11"}]},{"modelRef":"zai-org/GLM-4.6","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["terminal-bench"],"sourceIds":["hf-terminal-bench"],"sourceLabels":["Hugging Face · Terminal-Bench leaderboard API"],"sourceUrls":["https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","examples":[{"sourceId":"hf-terminal-bench","sourceLabel":"Hugging Face · Terminal-Bench leaderboard API","sourceUrl":"https://huggingface.co/api/datasets/harborframework/terminal-bench-2.0/leaderboard","evidenceUrl":"https://www.tbench.ai/leaderboard/terminal-bench/2.0","sourceLocator":"rank=37;row=36"}]},{"modelRef":"Gemini 3 Flash","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=47","observedAt":"2026-08-24"}]},{"modelRef":"Gemma 4 31B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=51","observedAt":"2026-08-24"}]},{"modelRef":"Grok 4.3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=50","observedAt":"2026-08-24"}]},{"modelRef":"Grok 4.3 (High)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=45","observedAt":"2026-08-24"}]},{"modelRef":"Grok Build 0.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=44","observedAt":"2026-08-24"}]},{"modelRef":"Hy3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=30","observedAt":"2026-08-24"}]},{"modelRef":"Inkling","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=42","observedAt":"2026-08-24"}]},{"modelRef":"Mistral Medium 3.5","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=40","observedAt":"2026-08-24"}]},{"modelRef":"Muse Spark 1.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=28","observedAt":"2026-08-24"}]},{"modelRef":"Muse Spark 1.2 (xHigh)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=25","observedAt":"2026-08-24"}]},{"modelRef":"Nemotron 3 Ultra","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=49","observedAt":"2026-08-24"}]},{"modelRef":"Solar Pro 4","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-agent"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=agent;split=latest;row_idx=43","observedAt":"2026-08-24"}]},{"modelRef":"api-gpt-4o-search","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-search"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=33","observedAt":"2026-08-24"}]},{"modelRef":"chatglm2-6b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=386","observedAt":"2026-08-21"}]},{"modelRef":"claude-opus-4-1-search","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-search"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=24","observedAt":"2026-08-24"}]},{"modelRef":"codellama-70b-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=364","observedAt":"2026-08-21"}]},{"modelRef":"diffbot-small-xl","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-search"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=32","observedAt":"2026-08-24"}]},{"modelRef":"dolphin-2.2.1-mistral-7b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=354","observedAt":"2026-08-21"}]},{"modelRef":"falcon-180b-chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=367","observedAt":"2026-08-21"}]},{"modelRef":"gemini-2.5-pro-grounding","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-search"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=26","observedAt":"2026-08-24"}]},{"modelRef":"gemini-3-flash-grounding","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-search"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=14","observedAt":"2026-08-24"}]},{"modelRef":"gemini-3-pro-grounding","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-search"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=9","observedAt":"2026-08-24"}]},{"modelRef":"gemini-3.1-pro-grounding","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-search"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=8","observedAt":"2026-08-24"}]},{"modelRef":"gemma-3-12b-it","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=200","observedAt":"2026-08-21"}]},{"modelRef":"gemma-3-4b-it","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=234","observedAt":"2026-08-21"}]},{"modelRef":"gpt4all-13b-snoozy","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=383","observedAt":"2026-08-21"}]},{"modelRef":"grok-4-1-fast-search","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-search"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=20","observedAt":"2026-08-24"}]},{"modelRef":"guanaco-33b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=369","observedAt":"2026-08-21"}]},{"modelRef":"llama-13b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=393","observedAt":"2026-08-21"}]},{"modelRef":"llama2-70b-steerlm-chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=344","observedAt":"2026-08-21"}]},{"modelRef":"mpt-30b-chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=360","observedAt":"2026-08-21"}]},{"modelRef":"nous-hermes-2-mixtral-8x7b-dpo","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=336","observedAt":"2026-08-21"}]},{"modelRef":"o3-search","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-search"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=25","observedAt":"2026-08-24"}]},{"modelRef":"ppl-sonar-pro-high","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-search"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=30","observedAt":"2026-08-24"}]},{"modelRef":"ppl-sonar-reasoning-pro-high","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-search"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-24","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=search;split=latest;row_idx=28","observedAt":"2026-08-24"}]},{"modelRef":"smollm2-1.7b-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=372","observedAt":"2026-08-21"}]},{"modelRef":"solar-10.7b-instruct-v1.0","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=353","observedAt":"2026-08-21"}]},{"modelRef":"stablelm-tuned-alpha-7b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=391","observedAt":"2026-08-21"}]},{"modelRef":"stripedhyena-nous-7b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=373","observedAt":"2026-08-21"}]},{"modelRef":"zephyr-7b-alpha","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["arena-text"],"sourceIds":["lmarena-hf-dataset"],"sourceLabels":["Arena · official Hugging Face leaderboard dataset"],"sourceUrls":["https://datasets-server.huggingface.co/rows"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T06:23:04Z","observedAtMax":"2026-08-21","examples":[{"sourceId":"lmarena-hf-dataset","sourceLabel":"Arena · official Hugging Face leaderboard dataset","sourceUrl":"https://datasets-server.huggingface.co/rows","evidenceUrl":"https://datasets-server.huggingface.co/rows","sourceLocator":"dataset=lmarena-ai/leaderboard-dataset;config=text;split=latest;row_idx=362","observedAt":"2026-08-21"}]},{"modelRef":"Codestral 25.01","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-01-13","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=391;dirname=2025-01-13-18-17-25--codestral-whole2","observedAt":"2025-01-13"}]},{"modelRef":"DeepSeek Chat V2.5","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-21","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=157;dirname=2024-12-21-20-56-21--polyglot-deepseek-diff","observedAt":"2024-12-21"}]},{"modelRef":"DeepSeek Chat V3 (prev)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-25","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=339;dirname=2024-12-25-13-31-51--deepseekv3preview-diff2","observedAt":"2024-12-25"}]},{"modelRef":"DeepSeek R1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-01-20","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=417;dirname=2025-01-20-19-11-38--ds-turns-upd-cur-msgs-fix-with-summarizer","observedAt":"2025-01-20"}]},{"modelRef":"DeepSeek R1 + claude-3-5-sonnet-20241022","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-01-23","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=443;dirname=2025-01-23-19-14-48--r1-architect-sonnet","observedAt":"2025-01-23"}]},{"modelRef":"DeepSeek V3 (0324)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-03-24","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=784;dirname=2025-03-24-15-41-33--deepseek-v3-0324-polyglot-diff","observedAt":"2025-03-24"}]},{"modelRef":"Gemini 2.0 Pro exp-02-05","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-02-25","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1;dirname=2025-02-25-20-23-07--gemini-pro","observedAt":"2025-02-25"}]},{"modelRef":"Gemini 2.5 Pro Preview 03-25","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-12","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=810;dirname=2025-04-12-04-55-50--gemini-25-pro-diff-fenced","observedAt":"2025-04-12"}]},{"modelRef":"Gemini 2.5 Pro Preview 05-06","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-05-07","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1174;dirname=2025-05-07-19-32-40--gemini0506-diff-fenced-completion_cost","observedAt":"2025-05-07"}]},{"modelRef":"Grok 3 Beta","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-10","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=914;dirname=2025-04-10-04-21-31--grok3-diff-exuser","observedAt":"2025-04-10"}]},{"modelRef":"Grok 3 Mini Beta (high)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-10","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=966;dirname=2025-04-10-23-59-02--xai-grok3-mini-whole-high","observedAt":"2025-04-10"}]},{"modelRef":"Grok 3 Mini Beta (low)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-10","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=940;dirname=2025-04-10-18-47-24--grok3-mini-whole-exuser","observedAt":"2025-04-10"}]},{"modelRef":"Kimi K2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-07-17","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1659;dirname=2025-07-17-17-41-54--kimi-k2-diff-or-pricing","observedAt":"2025-07-17"}]},{"modelRef":"Llama 4 Maverick","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-06","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=888;dirname=2025-04-06-08-39-52--llama-4-maverick-17b-128e-instruct-polyglot-whole","observedAt":"2025-04-06"}]},{"modelRef":"Optimus Alpha","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-10","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=992;dirname=2025-04-10-19-02-44--oalpha-diff-exsys","observedAt":"2025-04-10"}]},{"modelRef":"Quasar Alpha","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-04","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=862;dirname=2025-04-04-02-57-25--qalpha-diff-exsys","observedAt":"2025-04-04"}]},{"modelRef":"QwQ-32B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-03-06","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=678;dirname=2025-03-06-17-40-24--qwq32b-diff-temp-topp-ex-sys-remind-user-for-real","observedAt":"2025-03-06"}]},{"modelRef":"QwQ-32B + Qwen 2.5 Coder Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-03-07","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=704;dirname=2025-03-07-15-11-27--qwq32b-arch-temp-topp-again","observedAt":"2025-03-07"}]},{"modelRef":"Qwen3 235B A22B diff, no think, Alibaba API","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-05-09","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1228;dirname=2025-05-09-17-02-02--qwen3-235b-a22b.unthink_16k_diff","observedAt":"2025-05-09"}]},{"modelRef":"Qwen3 32B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-05-08","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1200;dirname=2025-05-08-03-20-24--qwen3-32b-default","observedAt":"2025-05-08"}]},{"modelRef":"chatgpt-4o-latest (2025-02-15)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-02-15","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=574;dirname=2025-02-15-19-51-22--chatgpt4o-feb15-diff","observedAt":"2025-02-15"}]},{"modelRef":"chatgpt-4o-latest (2025-03-29)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-03-29","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=836;dirname=2025-03-29-05-24-55--chatgpt4o-mar28-diff","observedAt":"2025-03-29"}]},{"modelRef":"claude-3-5-haiku-20241022","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-21","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=183;dirname=2024-12-21-21-46-27--polyglot-haiku-diff","observedAt":"2024-12-21"}]},{"modelRef":"claude-3-5-sonnet-20241022","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-01-17","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=53;dirname=2025-01-17-19-44-33--sonnet-baseline-jan-17","observedAt":"2025-01-17"}]},{"modelRef":"claude-3-7-sonnet-20250219 (32k thinking tokens)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-02-24","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=626;dirname=2025-02-24-21-47-23--sonnet37-diff-think-32k-64k","observedAt":"2025-02-24"}]},{"modelRef":"claude-3-7-sonnet-20250219 (no thinking)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-02-24","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=600;dirname=2025-02-24-19-54-07--sonnet37-diff","observedAt":"2025-02-24"}]},{"modelRef":"claude-sonnet-4-20250514 (32k thinking)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-05-24","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1284;dirname=2025-05-24-22-10-36--sonnet4-diff-exuser-think32k","observedAt":"2025-05-24"}]},{"modelRef":"claude-sonnet-4-20250514 (no thinking)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-05-24","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1256;dirname=2025-05-24-21-17-54--sonnet4-diff-exuser","observedAt":"2025-05-24"}]},{"modelRef":"command-a-03-2025-quality","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-03-14","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=732;dirname=2025-03-14-23-40-00--cmda-quality-whole2","observedAt":"2025-03-14"}]},{"modelRef":"gemini-2.0-flash-exp","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-22","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=287;dirname=2024-12-22-20-08-13--gemini-2.0-flash-exp-polyglot-whole","observedAt":"2024-12-22"}]},{"modelRef":"gemini-2.0-flash-thinking-exp-01-21","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-01-21","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=548;dirname=2025-01-21-22-51-49--gemini-2.0-flash-thinking-exp-01-21-polyglot-diff","observedAt":"2025-01-21"}]},{"modelRef":"gemini-2.5-flash-preview-04-17 (default)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-20","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1148;dirname=2025-04-20-19-54-31--flash25-diff-no-think","observedAt":"2025-04-20"}]},{"modelRef":"gemini-2.5-flash-preview-05-20 (24k think)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-05-25","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1399;dirname=2025-05-25-22-58-44--flash25-05-20-24k-think","observedAt":"2025-05-25"}]},{"modelRef":"gemini-2.5-flash-preview-05-20 (no think)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-05-26","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1370;dirname=2025-05-26-15-56-31--flash25-05-20-24k-think # dirname is misleading","observedAt":"2025-05-26"}]},{"modelRef":"gemini-2.5-pro-preview-06-05 (32k think)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-06-06","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1456;dirname=2025-06-06-16-36-21--gemini0605-32k-think-diff-fenced","observedAt":"2025-06-06"}]},{"modelRef":"gemini-2.5-pro-preview-06-05 (default think)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-06-06","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1428;dirname=2025-06-06-18-38-56--gemini0605-diff-fenced","observedAt":"2025-06-06"}]},{"modelRef":"gemini-exp-1206","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-22","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=261;dirname=2024-12-22-18-43-25--gemini-exp-1206-polyglot-whole-2","observedAt":"2024-12-22"}]},{"modelRef":"gemma-3-27b-it","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-03-15","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=758;dirname=2025-03-15-01-21-24--gemma3-27b-or","observedAt":"2025-03-15"}]},{"modelRef":"gpt-4.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-14","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1018;dirname=2025-04-14-21-05-54--gpt41-diff-exuser","observedAt":"2025-04-14"}]},{"modelRef":"gpt-4.1-mini","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-14","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1044;dirname=2025-04-14-21-27-53--gpt41mini-diff","observedAt":"2025-04-14"}]},{"modelRef":"gpt-4.1-nano","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-14","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1070;dirname=2025-04-14-22-46-01--gpt41nano-diff","observedAt":"2025-04-14"}]},{"modelRef":"gpt-4.5-preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-02-27","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=652;dirname=2025-02-27-20-26-15--gpt45-diff3","observedAt":"2025-02-27"}]},{"modelRef":"gpt-4o-2024-08-06","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-30","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=105;dirname=2024-12-30-20-44-54--gpt4o-ex-as-sys-clean-prompt","observedAt":"2024-12-30"}]},{"modelRef":"gpt-4o-2024-11-20","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-30","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=79;dirname=2024-12-30-20-57-12--gpt-4o-2024-11-20-ex-as-sys","observedAt":"2024-12-30"}]},{"modelRef":"gpt-4o-mini-2024-07-18","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-21","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=27;dirname=2024-12-21-18-41-18--polyglot-gpt-4o-mini","observedAt":"2024-12-21"}]},{"modelRef":"gpt-oss-120b (high)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-08-06","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1687;dirname=2025-08-06-04-54-48--gpt-oss-120b-high-polyglot","observedAt":"2025-08-06"}]},{"modelRef":"o1-2024-12-17 (high)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-21","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=131;dirname=2024-12-21-19-23-03--polyglot-o1-hard-diff","observedAt":"2024-12-21"}]},{"modelRef":"o1-mini-2024-09-12","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-22","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=235;dirname=2024-12-22-21-26-35--polyglot-o1mini-whole","observedAt":"2024-12-22"}]},{"modelRef":"o3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-06-25","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1542;dirname=2025-06-25-20-30-16--o3-price-reduction","observedAt":"2025-06-25"}]},{"modelRef":"o3 (high)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-06-25","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1513;dirname=2025-06-25-21-04-24--o3-price-reduction-high","observedAt":"2025-06-25"}]},{"modelRef":"o3 (high) + gpt-4.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-06-27","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1570;dirname=2025-06-27-23-53-57--o3-mini-high-diff-arch","observedAt":"2025-06-27"}]},{"modelRef":"o3-mini (high)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-01-31","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=522;dirname=2025-01-31-20-42-47--o3-mini-diff-high","observedAt":"2025-01-31"}]},{"modelRef":"o3-mini (medium)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-01-31","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=496;dirname=2025-01-31-20-27-46--o3-mini-diff2","observedAt":"2025-01-31"}]},{"modelRef":"o3-pro (high)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-06-28","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1601;dirname=2025-06-28-00-38-18--o3-pro-high","observedAt":"2025-06-28"}]},{"modelRef":"o4-mini (high)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-16","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1096;dirname=2025-04-16-22-01-58--o4-mini-high-diff-exsys","observedAt":"2025-04-16"}]},{"modelRef":"openhands-lm-32b-v0.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-04-19","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=1122;dirname=2025-04-19-14-43-04--o4-mini-patch","observedAt":"2025-04-19"}]},{"modelRef":"qwen-max-2025-01-25","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2025-01-28","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=471;dirname=2025-01-28-16-00-03--qwen-max-2025-01-25-polyglot-diff","observedAt":"2025-01-28"}]},{"modelRef":"yi-lightning","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-aider-polyglot"],"sourceLabels":["Aider · Polyglot coding leaderboard"],"sourceUrls":["https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","observedAtMax":"2024-12-23","examples":[{"sourceId":"src-aider-polyglot","sourceLabel":"Aider · Polyglot coding leaderboard","sourceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","evidenceUrl":"https://raw.githubusercontent.com/Aider-AI/aider/main/aider/website/_data/polyglot_leaderboard.yml","sourceLocator":"yaml-row=313;dirname=2024-12-23-01-11-56--yi-test","observedAt":"2024-12-23"}]},{"modelRef":"Amazon-Nova-2-Lite-v1:0","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=79;rank=80;model=Amazon-Nova-2-Lite-v1:0 (FC);mode=FC"}]},{"modelRef":"Amazon-Nova-Micro-v1:0","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=94;rank=95;model=Amazon-Nova-Micro-v1:0 (FC);mode=FC"}]},{"modelRef":"Amazon-Nova-Pro-v1:0","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=87;rank=88;model=Amazon-Nova-Pro-v1:0 (FC);mode=FC"}]},{"modelRef":"Arch-Agent-1.5B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=59;rank=60;model=Arch-Agent-1.5B;mode=unknown"}]},{"modelRef":"Arch-Agent-32B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=36;rank=37;model=Arch-Agent-32B;mode=unknown"}]},{"modelRef":"Arch-Agent-3B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=55;rank=56;model=Arch-Agent-3B;mode=unknown"}]},{"modelRef":"Bielik-11B-v2.3-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=98;rank=99;model=Bielik-11B-v2.3-Instruct (Prompt);mode=Prompt"}]},{"modelRef":"BitAgent-Bounty-8B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=35;rank=36;model=BitAgent-Bounty-8B;mode=unknown"}]},{"modelRef":"CoALM-70B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=73;rank=74;model=CoALM-70B;mode=unknown"}]},{"modelRef":"CoALM-8B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=83;rank=84;model=CoALM-8B;mode=unknown"}]},{"modelRef":"Command A","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=34;rank=35;model=Command A (FC);mode=FC"}]},{"modelRef":"Command A Reasoning","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=12;rank=13;model=Command A Reasoning (FC);mode=FC"}]},{"modelRef":"Command R7B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=60;rank=61;model=Command R7B (FC);mode=FC"}]},{"modelRef":"Falcon3-10B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=81;rank=82;model=Falcon3-10B-Instruct (FC);mode=FC"}]},{"modelRef":"Falcon3-1B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=105;rank=106;model=Falcon3-1B-Instruct (FC);mode=FC"}]},{"modelRef":"Falcon3-3B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=103;rank=104;model=Falcon3-3B-Instruct (FC);mode=FC"}]},{"modelRef":"Falcon3-7B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=90;rank=91;model=Falcon3-7B-Instruct (FC);mode=FC"}]},{"modelRef":"GLM-4.6","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=3;rank=4;model=GLM-4.6 (FC thinking);mode=FC thinking"}]},{"modelRef":"Gemma-3-12b-it","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=65;rank=66;model=Gemma-3-12b-it (Prompt);mode=Prompt"}]},{"modelRef":"Gemma-3-1b-it","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=108;rank=109;model=Gemma-3-1b-it (Prompt);mode=Prompt"}]},{"modelRef":"Gemma-3-27b-it","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=68;rank=69;model=Gemma-3-27b-it (Prompt);mode=Prompt"}]},{"modelRef":"Gemma-3-4b-it","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=100;rank=101;model=Gemma-3-4b-it (Prompt);mode=Prompt"}]},{"modelRef":"Granite-20b-FunctionCalling","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=92;rank=93;model=Granite-20b-FunctionCalling (FC);mode=FC"}]},{"modelRef":"Granite-3.1-8B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=80;rank=81;model=Granite-3.1-8B-Instruct (FC);mode=FC"}]},{"modelRef":"Granite-3.2-8B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=82;rank=83;model=Granite-3.2-8B-Instruct (FC);mode=FC"}]},{"modelRef":"Granite-4.0-350m","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=102;rank=103;model=Granite-4.0-350m (FC);mode=FC"}]},{"modelRef":"Grok-4-1-fast-non-reasoning","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=11;rank=12;model=Grok-4-1-fast-non-reasoning (FC);mode=FC"}]},{"modelRef":"Grok-4-1-fast-reasoning","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=4;rank=5;model=Grok-4-1-fast-reasoning (FC);mode=FC"}]},{"modelRef":"Hammer2.1-0.5b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=99;rank=100;model=Hammer2.1-0.5b (FC);mode=FC"}]},{"modelRef":"Hammer2.1-1.5b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=74;rank=75;model=Hammer2.1-1.5b (FC);mode=FC"}]},{"modelRef":"Hammer2.1-3b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=67;rank=68;model=Hammer2.1-3b (FC);mode=FC"}]},{"modelRef":"Hammer2.1-7b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=63;rank=64;model=Hammer2.1-7b (FC);mode=FC"}]},{"modelRef":"Llama-3.1-8B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=84;rank=85;model=Llama-3.1-8B-Instruct (Prompt);mode=Prompt"}]},{"modelRef":"Llama-3.1-Nemotron-Ultra-253B-v1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=107;rank=108;model=Llama-3.1-Nemotron-Ultra-253B-v1 (FC);mode=FC"}]},{"modelRef":"Llama-3.2-1B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=106;rank=107;model=Llama-3.2-1B-Instruct (FC);mode=FC"}]},{"modelRef":"Llama-3.2-3B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=97;rank=98;model=Llama-3.2-3B-Instruct (FC);mode=FC"}]},{"modelRef":"Llama-3.3-70B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=61;rank=62;model=Llama-3.3-70B-Instruct (FC);mode=FC"}]},{"modelRef":"Llama-4-Maverick-17B-128E-Instruct-FP8","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=49;rank=50;model=Llama-4-Maverick-17B-128E-Instruct-FP8 (FC);mode=FC"}]},{"modelRef":"Llama-4-Scout-17B-16E-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=71;rank=72;model=Llama-4-Scout-17B-16E-Instruct (FC);mode=FC"}]},{"modelRef":"MiniCPM3-4B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=96;rank=97;model=MiniCPM3-4B (Prompt);mode=Prompt"}]},{"modelRef":"MiniCPM3-4B-FC","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=85;rank=86;model=MiniCPM3-4B-FC (FC);mode=FC"}]},{"modelRef":"Ministral-8B-Instruct-2410","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=104;rank=105;model=Ministral-8B-Instruct-2410 (FC);mode=FC"}]},{"modelRef":"Mistral-Small-2506","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=58;rank=59;model=Mistral-Small-2506 (Prompt);mode=Prompt"}]},{"modelRef":"Mistral-small-2506","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=50;rank=51;model=Mistral-small-2506 (FC);mode=FC"}]},{"modelRef":"Moonshotai-Kimi-K2-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=10;rank=11;model=Moonshotai-Kimi-K2-Instruct (FC);mode=FC"}]},{"modelRef":"Nanbeige3.5-Pro-Thinking","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=31;rank=32;model=Nanbeige3.5-Pro-Thinking (FC);mode=FC"}]},{"modelRef":"Nanbeige4-3B-Thinking-2511","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=24;rank=25;model=Nanbeige4-3B-Thinking-2511 (FC);mode=FC"}]},{"modelRef":"Phi-4","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=69;rank=70;model=Phi-4 (Prompt);mode=Prompt"}]},{"modelRef":"Qwen3-1.7B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=70;rank=71;model=Qwen3-1.7B (FC);mode=FC"}]},{"modelRef":"RZN-T","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=95;rank=96;model=RZN-T (Prompt);mode=Prompt"}]},{"modelRef":"ToolACE-2-8B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=39;rank=40;model=ToolACE-2-8B (FC);mode=FC"}]},{"modelRef":"palmyra-x-004","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=75;rank=76;model=palmyra-x-004 (FC);mode=FC"}]},{"modelRef":"xLAM-2-1b-fc-r","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=64;rank=65;model=xLAM-2-1b-fc-r (FC);mode=FC"}]},{"modelRef":"xLAM-2-32b-fc-r","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=17;rank=18;model=xLAM-2-32b-fc-r (FC);mode=FC"}]},{"modelRef":"xLAM-2-3b-fc-r","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=41;rank=42;model=xLAM-2-3b-fc-r (FC);mode=FC"}]},{"modelRef":"xLAM-2-70b-fc-r","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=21;rank=22;model=xLAM-2-70b-fc-r (FC);mode=FC"}]},{"modelRef":"xLAM-2-8b-fc-r","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["bfcl"],"sourceIds":["src-bfcl"],"sourceLabels":["Berkeley Function Calling Leaderboard"],"sourceUrls":["https://gorilla.cs.berkeley.edu/data_overall.csv"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:26:35Z","examples":[{"sourceId":"src-bfcl","sourceLabel":"Berkeley Function Calling Leaderboard","sourceUrl":"https://gorilla.cs.berkeley.edu/data_overall.csv","evidenceUrl":"https://gorilla.cs.berkeley.edu/leaderboard","sourceLocator":"csv-row=33;rank=34;model=xLAM-2-8b-fc-r (FC);mode=FC"}]},{"modelRef":"DeepSeek-Coder-V2-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-gsm8k_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-06-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"gsm8k_external.csv:row=187;column=EM","observedAt":"2024-06-17"}]},{"modelRef":"DeepSeek-Coder-V2-Lite-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-gsm8k_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-06-13","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"gsm8k_external.csv:row=186;column=EM","observedAt":"2024-06-13"}]},{"modelRef":"DeepSeek-V2.5","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-21","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=59;column=Percent correct","observedAt":"2024-12-21"}]},{"modelRef":"Dracarys2-72B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-30","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://livebench.ai/#/","sourceLocator":"live_bench_external.csv:row=31;column=Global average","observedAt":"2024-09-30"}]},{"modelRef":"Dracarys2-Llama-3.1-70B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-08-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://livebench.ai/#/","sourceLocator":"live_bench_external.csv:row=43;column=Global average","observedAt":"2024-08-14"}]},{"modelRef":"Eurus-2-7B-PRIME","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-31","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=262;column=mean_score","observedAt":"2024-12-31"}]},{"modelRef":"InternVL-Chat-ViT-6B-Vicuna-13B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-08-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://proceedings.neurips.cc/paper_files/paper/2024/file/81a60d18e010b27b36cd465c6604b915-Paper-Conference.pdf","sourceLocator":"science_qa_external.csv:row=100;column=Score","observedAt":"2024-08-16"}]},{"modelRef":"InternVL-Chat-ViT-6B-Vicuna-7B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-12-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://proceedings.neurips.cc/paper_files/paper/2024/file/81a60d18e010b27b36cd465c6604b915-Paper-Conference.pdf","sourceLocator":"science_qa_external.csv:row=99;column=Score","observedAt":"2023-12-25"}]},{"modelRef":"LLaVA-Video-7B-Qwen2","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-mindcube_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mindcube_external.csv:row=4;column=Overall score","observedAt":"2024-09-02"}]},{"modelRef":"Llama-3.2-1B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-balrog_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=36;column=Average progress","observedAt":"2024-09-24"}]},{"modelRef":"Llama-3.2-3B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-balrog_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=34;column=Average progress","observedAt":"2024-09-24"}]},{"modelRef":"LongVA-7B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-mindcube_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-06-13","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mindcube_external.csv:row=5;column=Overall score","observedAt":"2024-06-13"}]},{"modelRef":"MM1-3B-Chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-03-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"science_qa_external.csv:row=84;column=Score","observedAt":"2024-03-14"}]},{"modelRef":"MM1-7B-Chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-03-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"science_qa_external.csv:row=85;column=Score","observedAt":"2024-03-14"}]},{"modelRef":"OLMo-2-1124-13B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-31","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://livebench.ai/#/","sourceLocator":"live_bench_external.csv:row=63;column=Global average","observedAt":"2024-12-31"}]},{"modelRef":"Phi-3.5-vision-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-08-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"science_qa_external.csv:row=83;column=Score","observedAt":"2024-08-16"}]},{"modelRef":"Pixtral-12B-2409","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-geobench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-09-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://geobench.org/","sourceLocator":"geobench_external.csv:row=33;column=ACW Avg Score","observedAt":"2024-09-17"}]},{"modelRef":"QwQ-32B (16K thinking)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-lech_mazur_writing_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://github.com/lechmazur/Writing","sourceLocator":"lech_mazur_writing_external.csv:row=7;column=Mean score","observedAt":"2025-03-05"}]},{"modelRef":"Qwen-VL-Max","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-spatialviz_bench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-01-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"spatialviz_bench_external.csv:row=9;column=Overall score","observedAt":"2024-01-18"}]},{"modelRef":"Qwen1.5-7B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/lite/latest/#/leaderboard/mmlu","sourceLocator":"mmlu_external.csv:row=213;column=EM","observedAt":"2024-02-04"}]},{"modelRef":"Qwen2-VL-72B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-balrog_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-08-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=32;column=Average progress","observedAt":"2024-08-29"}]},{"modelRef":"Qwen2-VL-7B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-balrog_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-08-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=37;column=Average progress","observedAt":"2024-08-29"}]},{"modelRef":"Qwen2.5-Coder-14B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-gsm8k_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-11-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"gsm8k_external.csv:row=190;column=EM","observedAt":"2024-11-06"}]},{"modelRef":"Qwen2.5-Coder-3B-Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-gsm8k_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-11-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2409.12186","sourceLocator":"gsm8k_external.csv:row=188;column=EM","observedAt":"2024-11-06"}]},{"modelRef":"Switch-Base","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-superglue_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2101.03961","sourceLocator":"superglue_external.csv:row=4;column=Score"}]},{"modelRef":"Switch-Large","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-superglue_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2101.03961","sourceLocator":"superglue_external.csv:row=6;column=Score"}]},{"modelRef":"Yi-large","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-05-13","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/lite/latest/#/leaderboard/mmlu","sourceLocator":"mmlu_external.csv:row=245;column=EM","observedAt":"2024-05-13"}]},{"modelRef":"blip2-opt-2.7b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-02-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://scienceqa.github.io/leaderboard.html","sourceLocator":"science_qa_external.csv:row=45;column=Score","observedAt":"2023-02-06"}]},{"modelRef":"c4ai-command-a-03-2025","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=52;column=Percent correct","observedAt":"2025-03-14"}]},{"modelRef":"chatgpt-4o-01-29-2025","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=61;column=Percent correct","observedAt":"2025-02-15"}]},{"modelRef":"chatgpt-4o-03-27-2025","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-29","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=29;column=Percent correct","observedAt":"2025-03-29"}]},{"modelRef":"chutes/DeepSeek-V3-0324","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-24","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=844;column=ECI Score","observedAt":"2025-03-24"}]},{"modelRef":"chutes/Gemma-3-27b-It","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-11","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=845;column=ECI Score","observedAt":"2025-03-11"}]},{"modelRef":"chutes/Llama-4-Scout-17B-16E Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-06","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=855;column=ECI Score","observedAt":"2025-04-06"}]},{"modelRef":"chutes/Qwen3-235B-A22B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=839;column=ECI Score","observedAt":"2025-04-28"}]},{"modelRef":"chutes/Qwen3-235B-A22B-Thinking-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-07-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=840;column=ECI Score","observedAt":"2025-07-25"}]},{"modelRef":"chutes/gpt-oss-120b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=835;column=ECI Score","observedAt":"2025-08-05"}]},{"modelRef":"chutes/gpt-oss-120b_high","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=836;column=ECI Score","observedAt":"2025-08-05"}]},{"modelRef":"claude-mythos-preview-early","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-metr_time_horizons_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-07","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"metr_time_horizons_external.csv:row=2;column=average_score","observedAt":"2026-04-07"}]},{"modelRef":"claude-opus-4-1-20250805_32K","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-08-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=175;column=ECI Score","observedAt":"2025-08-05"}]},{"modelRef":"code-davinci-002","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-gsm8k_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2022-03-15","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/classic/latest/#/groups/gsm","sourceLocator":"gsm8k_external.csv:row=3;column=EM","observedAt":"2022-03-15"}]},{"modelRef":"codestral-2501","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-01-13","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=62;column=Percent correct","observedAt":"2025-01-13"}]},{"modelRef":"codestral-2508","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-ale_bench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-07-30","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=111;column=Performance","observedAt":"2025-07-30"}]},{"modelRef":"davinci-002","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-metr_time_horizons_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/","sourceLocator":"metr_time_horizons_external.csv:row=50;column=average_score"}]},{"modelRef":"falcon-11B-vlm","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-05-21","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2407.14885","sourceLocator":"science_qa_external.csv:row=92;column=Score","observedAt":"2024-05-21"}]},{"modelRef":"gdm-ai-co-mathematician","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-frontiermath_tier_4"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-05-08","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=10;column=mean_score","observedAt":"2026-05-08"}]},{"modelRef":"gemini-1.0-pro-vision","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-01-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"science_qa_external.csv:row=90;column=Score","observedAt":"2024-01-04"}]},{"modelRef":"gemini-2.0-flash-lite","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://livebench.ai/#/","sourceLocator":"live_bench_external.csv:row=29;column=Global average","observedAt":"2025-02-05"}]},{"modelRef":"gemini-2.0-flash-lite-001","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-forecastbench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=69;column=Overall score","observedAt":"2025-02-25"}]},{"modelRef":"gemini-2.0-flash-lite-preview-02-05","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://livebench.ai/#/","sourceLocator":"live_bench_external.csv:row=30;column=Global average","observedAt":"2025-02-05"}]},{"modelRef":"gemini-2.5-flash-lite-preview-06-17_16K","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-weirdml_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-06-17","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://htihle.github.io/weirdml.html","sourceLocator":"weirdml_external.csv:row=134;column=Accuracy","observedAt":"2025-06-17"}]},{"modelRef":"gemini-2.5-flash-lite-preview-09-2025","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-vpct_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://cbrower.dev/vpct","sourceLocator":"vpct_external.csv:row=39;column=Correct","observedAt":"2025-09-25"}]},{"modelRef":"gemini-robotics-er-1.5-preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-vpct_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-26","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://cbrower.dev/vpct","sourceLocator":"vpct_external.csv:row=16;column=Correct","observedAt":"2025-09-26"}]},{"modelRef":"gemini-robotics-er-1.6-preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-blueprint_bench_2_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-14","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"blueprint_bench_2_external.csv:row=22;column=Score","observedAt":"2026-04-14"}]},{"modelRef":"gpt-3.5-turbo-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-metr_time_horizons_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-09-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/","sourceLocator":"metr_time_horizons_external.csv:row=49;column=average_score","observedAt":"2023-09-18"}]},{"modelRef":"grok-3-mini-beta_medium","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-09","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=847;column=ECI Score","observedAt":"2025-04-09"}]},{"modelRef":"grok-4-1-fast-non-reasoning","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-simplebench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-11-19","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"simplebench_external.csv:row=41;column=Score (AVG@5)","observedAt":"2025-11-19"}]},{"modelRef":"grok-4-heavy-web-app","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-frontiermath_tier_4"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-07-10","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath_tier_4.csv:row=52;column=mean_score","observedAt":"2025-07-10"}]},{"modelRef":"instructblip-vicuna-13b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-12-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://proceedings.neurips.cc/paper_files/paper/2024/file/81a60d18e010b27b36cd465c6604b915-Paper-Conference.pdf","sourceLocator":"science_qa_external.csv:row=97;column=Score","observedAt":"2023-12-25"}]},{"modelRef":"instructblip-vicuna-7b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-05-22","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://proceedings.neurips.cc/paper_files/paper/2024/file/81a60d18e010b27b36cd465c6604b915-Paper-Conference.pdf","sourceLocator":"science_qa_external.csv:row=96;column=Score","observedAt":"2023-05-22"}]},{"modelRef":"learnlm-1.5-pro-experimental","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://livebench.ai/#/","sourceLocator":"live_bench_external.csv:row=34;column=Global average"}]},{"modelRef":"llama3-llava-next-8b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-20","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2404.14219","sourceLocator":"science_qa_external.csv:row=87;column=Score","observedAt":"2024-04-20"}]},{"modelRef":"llava-v1.5-7b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2023-10-05","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://proceedings.neurips.cc/paper_files/paper/2024/file/81a60d18e010b27b36cd465c6604b915-Paper-Conference.pdf","sourceLocator":"science_qa_external.csv:row=101;column=Score","observedAt":"2023-10-05"}]},{"modelRef":"llava-v1.6-mistral-7b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-01-31","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2407.14885","sourceLocator":"science_qa_external.csv:row=95;column=Score","observedAt":"2024-01-31"}]},{"modelRef":"llava-v1.6-vicuna-13b","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-science_qa_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-01-31","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2407.14885","sourceLocator":"science_qa_external.csv:row=94;column=Score","observedAt":"2024-01-31"}]},{"modelRef":"mPLUG-Owl3-7B-241101","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-mindcube_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-11-26","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"mindcube_external.csv:row=3;column=Overall score","observedAt":"2024-11-26"}]},{"modelRef":"mistral-small-2402","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-26","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/lite/latest/#/leaderboard/mmlu","sourceLocator":"mmlu_external.csv:row=168;column=EM","observedAt":"2024-02-26"}]},{"modelRef":"mistral-small-2603","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-ale_bench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-03-16","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=91;column=Performance","observedAt":"2026-03-16"}]},{"modelRef":"nova-2.0-pro-preview_unknown","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-gdp_pdf_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-12-02","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://surgehq.ai/benchmarks/gdp-pdf","sourceLocator":"gdp_pdf_external.csv:row=28;column=GDP.pdf score","observedAt":"2025-12-02"}]},{"modelRef":"o1-pro-2025-03-19_low","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-arc_agi_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-19","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://arcprize.org/leaderboard","sourceLocator":"arc_agi_external.csv:row=190;column=Score","observedAt":"2025-03-19"}]},{"modelRef":"openhands-lm-32b-v0.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-19","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=12;column=Percent correct","observedAt":"2025-04-19"}]},{"modelRef":"parasail-qwen3-235b-a22b-instruct-2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-epoch_capabilities_index"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-01","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"epoch_capabilities_index.csv:row=850;column=ECI Score","observedAt":"2025-09-01"}]},{"modelRef":"pixtral-large-2411","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-enigma_eval_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-11-18","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"enigma_eval_external.csv:row=37;column=Accuracy","observedAt":"2024-11-18"}]},{"modelRef":"qwen-plus-2025-04-28","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["frontiermath"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-28","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiermath.csv:row=73;column=mean_score","observedAt":"2025-04-28"}]},{"modelRef":"qwen1.5-110b-chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-forecastbench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-25","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"forecastbench_external.csv:row=62;column=Overall score","observedAt":"2024-04-25"}]},{"modelRef":"qwen1.5-14B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/lite/latest/#/leaderboard/mmlu","sourceLocator":"mmlu_external.csv:row=214;column=EM","observedAt":"2024-02-04"}]},{"modelRef":"qwen1.5-32B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://crfm.stanford.edu/helm/lite/latest/#/leaderboard/mmlu","sourceLocator":"mmlu_external.csv:row=215;column=EM","observedAt":"2024-02-04"}]},{"modelRef":"qwen1.5-32b-chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-04-03","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=240;column=mean_score","observedAt":"2024-04-03"}]},{"modelRef":"qwen1.5-72b-chat","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-02-04","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=239;column=mean_score","observedAt":"2024-02-04"}]},{"modelRef":"qwen2.5-14b-instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-mmlu_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-02-26","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"http://arxiv.org/abs/2412.08905","sourceLocator":"mmlu_external.csv:row=219;column=EM","observedAt":"2025-02-26"}]},{"modelRef":"qwen3-coder-plus","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-ale_bench_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-09-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"ale_bench_external.csv:row=93;column=Performance","observedAt":"2025-09-23"}]},{"modelRef":"qwq-plus","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["gpqa-diamond"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-04-08","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"gpqa_diamond.csv:row=177;column=mean_score","observedAt":"2025-04-08"}]},{"modelRef":"reka-flash-3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-balrog_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2025-03-10","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://balrogai.com/","sourceLocator":"balrog_external.csv:row=19;column=Average progress","observedAt":"2025-03-10"}]},{"modelRef":"sonar","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["livebench"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://livebench.ai/#/","sourceLocator":"live_bench_external.csv:row=41;column=Global average"}]},{"modelRef":"swe-1.6","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-frontiercode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-04-07","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=33;column=Main score","observedAt":"2026-04-07"}]},{"modelRef":"swe-1.7","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["epoch-frontiercode_external"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2026-07-08","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://epoch.ai/data/benchmark_data.zip","sourceLocator":"frontiercode_external.csv:row=13;column=Main score","observedAt":"2026-07-08"}]},{"modelRef":"yi-lightning","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["aider-polyglot"],"sourceIds":["src-epoch-benchmark-hub"],"sourceLabels":["Epoch AI · Benchmarking Hub downloadable snapshot"],"sourceUrls":["https://epoch.ai/data/benchmark_data.zip"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T08:25:00Z","observedAtMax":"2024-12-23","examples":[{"sourceId":"src-epoch-benchmark-hub","sourceLabel":"Epoch AI · Benchmarking Hub downloadable snapshot","sourceUrl":"https://epoch.ai/data/benchmark_data.zip","evidenceUrl":"https://aider.chat/docs/leaderboards/#polyglot-leaderboard","sourceLocator":"aider_polyglot_external.csv:row=65;column=Percent correct","observedAt":"2024-12-23"}]},{"modelRef":"4x Scaled","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-02-03","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[74]","observedAt":"2025-02-03"}]},{"modelRef":"Amazon.nova Premier v1:0","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-05-27","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[126]","observedAt":"2025-05-27"}]},{"modelRef":"Claude 3.5 Haiku","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-10-22","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[134]","observedAt":"2024-10-22"}]},{"modelRef":"Claude 3.7 Sonnet w/ Review Heavy","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-02-25","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[71]","observedAt":"2025-02-25"}]},{"modelRef":"Co-PatcheR","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-05-28","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[118]","observedAt":"2025-05-28"}]},{"modelRef":"CodeAct v1.8","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-lite"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-07-25","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[60]","observedAt":"2024-07-25"}]},{"modelRef":"DeepSWE-Preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-06-29","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[128]","observedAt":"2025-06-29"}]},{"modelRef":"DeepSeek V3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-lite"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-06-09","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[35]","observedAt":"2025-06-09"}]},{"modelRef":"DeepSeek V3 0324","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-08-06","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[129]","observedAt":"2025-08-06"}]},{"modelRef":"Deepseek V3","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-lite"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-01-11","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[47]","observedAt":"2025-01-11"}]},{"modelRef":"DevStral Small 2505","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-05-20","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[115]","observedAt":"2025-05-20"}]},{"modelRef":"DevStral Small 2507","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-07-25","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[144]","observedAt":"2025-07-25"}]},{"modelRef":"Doubao-Seed-Code","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-09-28","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[2]","observedAt":"2025-09-28"}]},{"modelRef":"Frogboss 32B 2510","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-11-10","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[96]","observedAt":"2025-11-10"}]},{"modelRef":"Frogmini 14B 2510","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-11-10","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[121]","observedAt":"2025-11-10"}]},{"modelRef":"GLM-4.5","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-07-28","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[65]","observedAt":"2025-07-28"}]},{"modelRef":"GLM-4.6","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-09-30","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[50]","observedAt":"2025-09-30"}]},{"modelRef":"GPT-4 (0125)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-lite"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-05-30","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[73]","observedAt":"2024-05-30"}]},{"modelRef":"GPT-4 (0613)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-lite"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-05-24","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[67]","observedAt":"2024-05-24"}]},{"modelRef":"GPT-4 (0806)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-lite"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-01-13","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[28]","observedAt":"2025-01-13"}]},{"modelRef":"GPT-4.1","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-multimodal"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-05-31","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[5]=Multimodal;results[6]","observedAt":"2025-05-31"}]},{"modelRef":"GPT-4o & Claude 3 Opus","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-lite"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-05-23","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[62]","observedAt":"2024-05-23"}]},{"modelRef":"Gemini 2.0 Flash (Experimental)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-01-18","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[123]","observedAt":"2025-01-18"}]},{"modelRef":"Gemini 2.0 Flash (v20241212-experimental)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-12-12","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[103]","observedAt":"2024-12-12"}]},{"modelRef":"Lingma SWE-GPT 72b (v0918)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-09-18","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[160]","observedAt":"2024-09-18"}]},{"modelRef":"Lingma SWE-GPT 72b (v0925)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-10-02","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[153]","observedAt":"2024-10-02"}]},{"modelRef":"Lingma SWE-GPT 7b (v0918)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-09-18","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[171]","observedAt":"2024-09-18"}]},{"modelRef":"Lingma SWE-GPT 7b (v0925)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-10-02","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[168]","observedAt":"2024-10-02"}]},{"modelRef":"Llama3-SWE-RL-70B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-02-26","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[132]","observedAt":"2025-02-26"}]},{"modelRef":"Mixed Models","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-lite"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-07-02","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[20]","observedAt":"2024-07-02"}]},{"modelRef":"Qwen2.5 Coder 32B Instruct","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-06-16","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[143]","observedAt":"2025-06-16"}]},{"modelRef":"SWE-agent-LM-32B","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-05-11","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[137]","observedAt":"2025-05-11"}]},{"modelRef":"TTS(Bo16)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-06-29","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[79]","observedAt":"2025-06-29"}]},{"modelRef":"TTS(Bo8)","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-06-16","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[114]","observedAt":"2025-06-16"}]},{"modelRef":"claude-3-5-sonnet-20241022","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-lite"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-12-03","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[51]","observedAt":"2024-12-03"}]},{"modelRef":"o1-preview","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-01-17","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[63]","observedAt":"2025-01-17"}]},{"modelRef":"o3-mini_1.0","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-lite"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2025-02-07","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[4]=Lite;results[49]","observedAt":"2025-02-07"}]},{"modelRef":"swe-search","rowCount":1,"numericRowCount":1,"benchmarkCount":1,"benchmarkIds":["swebench-verified"],"sourceIds":["swebench-official"],"sourceLabels":["SWE-bench · official leaderboard JSON"],"sourceUrls":["https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json"],"mappingStatusCounts":{"unmatched":1},"latestRetrievedAt":"2026-08-27T04:57:09Z","observedAtMax":"2024-12-21","examples":[{"sourceId":"swebench-official","sourceLabel":"SWE-bench · official leaderboard JSON","sourceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","evidenceUrl":"https://raw.githubusercontent.com/swe-bench/swe-bench.github.io/master/data/leaderboards.json","sourceLocator":"leaderboards[3]=Verified;results[72]","observedAt":"2024-12-21"}]}],"publicMeta":{"aliasRegistry":"data/public/model_aliases.json","errors":[],"generatedAt":"2026-08-27T16:21:44Z","inputFiles":["swebench-official/candidates.jsonl","src-epoch-benchmark-hub/candidates.jsonl","src-aider-polyglot/candidates.jsonl","src-bfcl/candidates.jsonl","src-mle-bench/candidates.jsonl","agents-last-exam/candidates.jsonl","helm-capabilities/candidates.jsonl","helm-lite/candidates.jsonl","hf-aime-2026/candidates.jsonl","hf-gpqa/candidates.jsonl","hf-hle/candidates.jsonl","hf-hmmt-2026/candidates.jsonl","hf-mmlu-pro/candidates.jsonl","hf-swebench-pro/candidates.jsonl","hf-swebench-verified/candidates.jsonl","hf-terminal-bench/candidates.jsonl","livebench-official/candidates.jsonl","lmarena-hf-dataset/candidates.jsonl"],"note":"公开榜单/模型卡报告值的展示层；不等同于本项目复现，也不覆盖 canonical observations。","payloadHashes":["0e951886a0a40b63e3d2b160988c29c21fa72eab5e447daedfab67c2323a3121","2a0ab0a3296a218e1531287ca7eeab13d7863d305a00987333f0880bfa506ae9","2e159d9614de10ac2db4a6f240abde1aa936df4b4a22fe613d61729786f322b9","30a9286f868f3414905cbfbf3606917e5c384a3448e17df957bccaa1eeb7a602","4ad809bc2138e22252e0c8b2501981ee2661b34be7bee9b66b209ab86264bcab","586bbc19cb2a3e11be61021866ef6436cf4b4d935bb1db088984b567b76595bb","63e1661cf2158549e56da402bec46d0be439959a8db8a78603da28d3543bc04c","849df1ab0c91cce689080bcd3bc3612b64f8fb23e5290b7db80be509dacfc931","85a50b25953512d18ba4bb0c23c0b8e626fcf9a5b52d287644b8a0b44b9535de","9d2db66b8b3c418efe70744b36f6311fed461ad78c951cbc96159cb7ab400ff8","a58848d66cc6d2949246bbf78edc8327e8611019db601e3603a2238da1ee2b74","b4216be476f616e31837b47784db8c4591628bec33ffdfb2cc2f3ea96b924557","b918fd3f4cc409b820f49e53234b47d024663c1810bd3b1e6acfcfa3bf220a32","bd8f3e47321d266bfd4c817736e5f0d18cbcd5500ebef2c3b98cffdf31777d0c","d5b2c8331e24dc8662241e174033056037f05b642347b3453c24781a72578b2b","e2ead78a77de0e7c2859ae79d758ed9de7c73cbc5bd4d935ac810905589bb1fb","ede5081a24bc5ac7f9b3b09a52af6bcfcd35aa0f1709239fed979db77a8159a6","fa4b61d3167dfe99e1a834e007a38372c5bac07b7627f8e2c3904fb48cd4a006"],"reviewStatus":"unreviewed","schemaVersion":"public-evidence@0.1","selection":{"maxPerModelBenchmarkMetricSource":0,"omittedPolicy":"complete omitted rows are written to unmapped/alternatives JSONL","selectedPolicy":"all mapped rows by default; positive max_per_key keeps the highest evidence/current/featured/numeric variants"},"status":"reported-unverified","verified":false},"publicStats":{"benchmarkCounts":{"agents-last-exam":1426,"aider-polyglot":141,"aime-2026":24,"arena-agent":52,"arena-document":38,"arena-search":34,"arena-text":1000,"arena-vision":987,"arena-webdev":499,"bfcl":109,"epoch-adversarial_nli_external":15,"epoch-ale_bench_external":110,"epoch-algotune_external":18,"epoch-apex_agents_external":62,"epoch-arc_agi_2_external":216,"epoch-arc_agi_external":233,"epoch-arc_ai2_external":134,"epoch-balrog_external":36,"epoch-bbh_external":88,"epoch-blueprint_bench_2_external":23,"epoch-bool_q_external":136,"epoch-btf3_external":10,"epoch-chess_puzzles":163,"epoch-cl_bench_external":23,"epoch-cl_bench_life_external":17,"epoch-common_sense_qa_2_external":6,"epoch-critpt_external":139,"epoch-cursorbench_external":67,"epoch-deepresearchbench_external":41,"epoch-deepswe_external":61,"epoch-enigma_eval_external":46,"epoch-epoch_capabilities_index":568,"epoch-forecastbench_external":81,"epoch-frontiercode_external":28,"epoch-frontiermath_tier_4":72,"epoch-frontierswe_external":17,"epoch-gbaeval_external":23,"epoch-gdp_pdf_external":26,"epoch-gdpval_external":11,"epoch-geobench_external":32,"epoch-gsm8k_external":170,"epoch-gso_external":38,"epoch-hella_swag_external":113,"epoch-lambada_external":53,"epoch-lech_mazur_writing_external":49,"epoch-math_level_5":108,"epoch-metr_time_horizons_external":50,"epoch-mindcube_external":5,"epoch-mirrorcode":6,"epoch-mmlu_external":217,"epoch-mystery_game_puzzles":54,"epoch-open_book_qa_external":70,"epoch-osworld_2_external":10,"epoch-otis_mock_aime_2024_2025":240,"epoch-piqa_external":113,"epoch-proofbench_external":59,"epoch-rli_external":13,"epoch-scicode_external":129,"epoch-science_qa_external":26,"epoch-simplebench_external":102,"epoch-spatialviz_bench_external":8,"epoch-superglue_external":10,"epoch-surface_evolver_bench_external":25,"epoch-terminalbench_external":202,"epoch-the_agent_company_external":16,"epoch-trivia_qa_external":115,"epoch-vending_bench_2_external":59,"epoch-vpct_external":38,"epoch-webdev_arena_external":117,"epoch-weirdml_external":166,"epoch-wino_grande_external":137,"frontiermath":101,"gpqa-diamond":380,"helm-gpqa":476,"helm-gsm8k":637,"helm-ifeval":476,"helm-legalbench":637,"helm-math":637,"helm-mean-score":136,"helm-mean-win-rate":273,"helm-medqa":637,"helm-mmlu":637,"helm-mmlu-pro":476,"helm-narrativeqa":637,"helm-naturalquestions-closed-book":637,"helm-naturalquestions-open-book":637,"helm-omni-math":476,"helm-openbookqa":637,"helm-wildbench":476,"helm-wmt-2014":637,"hle":141,"hmmt":17,"livebench":54,"livebench-amps-hard":49,"livebench-code-completion":49,"livebench-code-generation":49,"livebench-connections":49,"livebench-consecutive-events":49,"livebench-integrals-with-game":49,"livebench-javascript":49,"livebench-logic-with-navigation":49,"livebench-math-comp":49,"livebench-olympiad":49,"livebench-paraphrase":49,"livebench-plot-unscrambling":49,"livebench-python":49,"livebench-simplify":49,"livebench-spatial":49,"livebench-story-generation":49,"livebench-summarize":49,"livebench-tablejoin":49,"livebench-tablereformat":49,"livebench-theory-of-mind":49,"livebench-typescript":49,"livebench-typos":49,"livebench-zebra-puzzle":49,"mle-bench":112,"mmlu-pro":143,"osworld":20,"simpleqa":79,"swebench-bash-only":47,"swebench-lite":84,"swebench-multilingual":13,"swebench-multimodal":22,"swebench-pro":42,"swebench-test":24,"swebench-verified":292,"terminal-bench":40},"benchmarks":128,"candidateRows":98,"conflictRows":0,"curatedAliasRows":2830,"deduplicatedRows":21227,"harnesses":225,"inputRows":21227,"mappedCells":2750,"mappedMetricCells":3064,"mappedPerformanceMetricCells":2824,"mappedPerformanceRows":6191,"mappedRows":6461,"mappedTelemetryMetricCells":240,"mappedTelemetryRows":270,"mappingCounts":{"curated_alias":2830,"exact_alias":1570,"heuristic_alias":2061,"unmatched":14766},"models":1391,"numericRows":21129,"omittedRows":14766,"reportedRows":21129,"retrievedAtMax":"2026-08-27T08:26:35Z","retrievedAtMin":"2026-08-27T04:57:09Z","rows":6461,"selectedCandidateRows":0,"selectedReportedRows":6461,"selectedRows":6461,"snapshotCount":21227,"sourceCounts":{"agents-last-exam":1426,"helm-capabilities":2516,"helm-lite":6643,"hf-aime-2026":24,"hf-gpqa":115,"hf-hle":90,"hf-hmmt-2026":17,"hf-mmlu-pro":143,"hf-swebench-pro":42,"hf-swebench-verified":77,"hf-terminal-bench":40,"livebench-official":1127,"lmarena-hf-dataset":2610,"src-aider-polyglot":69,"src-bfcl":109,"src-epoch-benchmark-hub":5697,"src-mle-bench":112,"swebench-official":370},"sources":18,"statusCounts":{"candidate":98,"reported":21129},"unmappedRows":14766}}